WEBVTT 00:00:00.000 --> 00:00:09.337 align:middle line:84% [SQUEAKING] [RUSTLING] [CLICKING] 00:00:09.337 --> 00:00:10.420 align:middle line:90% JOSH MCDERMOTT: All right. 00:00:10.420 --> 00:00:14.040 align:middle line:84% So today, we will continue our journey 00:00:14.040 --> 00:00:15.640 align:middle line:90% into the auditory system. 00:00:15.640 --> 00:00:18.240 align:middle line:84% Last time, we talked about the ear. 00:00:18.240 --> 00:00:20.938 align:middle line:84% And this time, we're going to be talking about one 00:00:20.938 --> 00:00:23.480 align:middle line:84% of the most important properties of the ear, namely frequency 00:00:23.480 --> 00:00:24.930 align:middle line:90% selective as well as the-- 00:00:24.930 --> 00:00:26.680 align:middle line:84% frequency selectivity, as well as the fact 00:00:26.680 --> 00:00:28.840 align:middle line:90% that the ear is nonlinear. 00:00:28.840 --> 00:00:31.840 align:middle line:84% And then we're going to get into hearing loss, which 00:00:31.840 --> 00:00:34.960 align:middle line:84% will probably affect all of you at some point in your life. 00:00:34.960 --> 00:00:37.080 align:middle line:84% So just a refresher, this is the ear. 00:00:37.080 --> 00:00:40.780 align:middle line:84% Remember, sound enters here, travels down the ear canal, 00:00:40.780 --> 00:00:42.980 align:middle line:84% causes the eardrum to vibrate back and forth. 00:00:42.980 --> 00:00:45.760 align:middle line:84% Those vibrations are transmitted via the ossicles 00:00:45.760 --> 00:00:48.480 align:middle line:84% through to the cochlea, which is the organ that 00:00:48.480 --> 00:00:50.480 align:middle line:84% does the sensory transduction for hearing. 00:00:50.480 --> 00:00:52.360 align:middle line:84% And so remember, we think of the ear 00:00:52.360 --> 00:00:55.300 align:middle line:84% as having these three main functional components. 00:00:55.300 --> 00:00:58.400 align:middle line:84% There's the outer ear, the pinna and the eardrum. 00:00:58.400 --> 00:01:00.730 align:middle line:84% There's the middle ear, which serves these two 00:01:00.730 --> 00:01:03.250 align:middle line:84% main functions-- impedance matching, which is necessary 00:01:03.250 --> 00:01:06.355 align:middle line:84% because you go from air to fluid, and overload protection. 00:01:06.355 --> 00:01:07.730 align:middle line:84% Remember, there are those muscles 00:01:07.730 --> 00:01:10.290 align:middle line:84% on the bones of the middle ear that 00:01:10.290 --> 00:01:13.553 align:middle line:84% can contract in the presence of high intensity sound. 00:01:13.553 --> 00:01:14.970 align:middle line:84% And then we spent most of our time 00:01:14.970 --> 00:01:16.730 align:middle line:84% last time talking about the inner ear. 00:01:16.730 --> 00:01:19.147 align:middle line:84% And we think of one of the main functions of the inner ear 00:01:19.147 --> 00:01:21.730 align:middle line:90% as being frequency analysis. 00:01:21.730 --> 00:01:24.930 align:middle line:84% And here are some numbers that are worth knowing 00:01:24.930 --> 00:01:27.090 align:middle line:90% and fun to think about. 00:01:27.090 --> 00:01:32.650 align:middle line:84% So in each of your ears, there are approximately 3,500 inner 00:01:32.650 --> 00:01:33.690 align:middle line:90% hair cells. 00:01:33.690 --> 00:01:36.290 align:middle line:84% Remember, the inner hair cells are the cells 00:01:36.290 --> 00:01:38.350 align:middle line:90% that move in response to sound. 00:01:38.350 --> 00:01:41.730 align:middle line:84% And when they are moved, their membrane potential changes. 00:01:41.730 --> 00:01:46.730 align:middle line:84% So they transduce mechanical energy into a voltage. 00:01:46.730 --> 00:01:50.390 align:middle line:84% There are about 12,000 outer hair cells. 00:01:50.390 --> 00:01:52.690 align:middle line:84% So remember that, typically, you have 00:01:52.690 --> 00:01:54.730 align:middle line:84% these three rows of outer hair cells, 00:01:54.730 --> 00:01:56.950 align:middle line:84% and then one parallel row of inner hair cells. 00:01:56.950 --> 00:01:58.980 align:middle line:90% So there's about a 3 to 1 ratio. 00:01:58.980 --> 00:02:01.260 align:middle line:84% And the outer hair cells are like the inner hair 00:02:01.260 --> 00:02:02.500 align:middle line:90% cells in reverse. 00:02:02.500 --> 00:02:04.260 align:middle line:84% So they get an electrical signal. 00:02:04.260 --> 00:02:06.440 align:middle line:84% And that causes their shape to change. 00:02:06.440 --> 00:02:08.580 align:middle line:84% Remember, we saw the video of the outer hair 00:02:08.580 --> 00:02:12.380 align:middle line:84% cell that was being supplied an electrical signal, which 00:02:12.380 --> 00:02:13.120 align:middle line:90% was music. 00:02:13.120 --> 00:02:14.980 align:middle line:84% And you saw the outer hair cell dancing. 00:02:14.980 --> 00:02:17.905 align:middle line:90% 00:02:17.905 --> 00:02:21.860 align:middle line:84% The auditory nerve fibers are these essentially wires 00:02:21.860 --> 00:02:24.620 align:middle line:84% that are carrying information from the inner hair 00:02:24.620 --> 00:02:26.020 align:middle line:90% cells to the brain. 00:02:26.020 --> 00:02:29.700 align:middle line:84% And there's about 30,000 auditory nerve fibers per ear, 00:02:29.700 --> 00:02:34.563 align:middle line:84% so roughly about 10 per inner hair cell. 00:02:34.563 --> 00:02:35.980 align:middle line:84% So just for comparison-- and we'll 00:02:35.980 --> 00:02:39.120 align:middle line:84% talk a lot more about the visual system later in the class. 00:02:39.120 --> 00:02:42.580 align:middle line:84% But these are analogous numbers in each of your eyes. 00:02:42.580 --> 00:02:45.200 align:middle line:84% So there are two types of photoreceptors in the eye. 00:02:45.200 --> 00:02:47.540 align:middle line:90% There are cones and rods. 00:02:47.540 --> 00:02:50.560 align:middle line:84% The cones, there are about 5 million of those. 00:02:50.560 --> 00:02:53.040 align:middle line:84% There's far more of the rods, 100 million. 00:02:53.040 --> 00:02:55.470 align:middle line:84% The rods typically function in conditions 00:02:55.470 --> 00:02:57.050 align:middle line:84% where there are low light levels. 00:02:57.050 --> 00:02:59.430 align:middle line:84% But the interesting discrepancy or difference 00:02:59.430 --> 00:03:01.430 align:middle line:84% with respect to the auditory system 00:03:01.430 --> 00:03:05.390 align:middle line:84% is that there are actually fewer optic nerve fibers 00:03:05.390 --> 00:03:08.950 align:middle line:84% than there are receptors, whereas, in the auditory system, 00:03:08.950 --> 00:03:10.690 align:middle line:90% it's the opposite. 00:03:10.690 --> 00:03:14.110 align:middle line:84% So in the visual system, you're taking the super high bandwidth 00:03:14.110 --> 00:03:17.743 align:middle line:84% signal and squeezing it down into a narrow pipe. 00:03:17.743 --> 00:03:19.410 align:middle line:84% And that's what gets sent to your brain. 00:03:19.410 --> 00:03:21.118 align:middle line:84% And there's probably lots of consequences 00:03:21.118 --> 00:03:24.370 align:middle line:90% for the neural coding from that. 00:03:24.370 --> 00:03:27.770 align:middle line:84% And in the auditory system, it's just the opposite relationship. 00:03:27.770 --> 00:03:30.930 align:middle line:84% And we can talk in more detail later on about why that is. 00:03:30.930 --> 00:03:35.790 align:middle line:84% It's interesting to think about, but it's a salient difference. 00:03:35.790 --> 00:03:38.870 align:middle line:84% So one of the most important attributes 00:03:38.870 --> 00:03:41.770 align:middle line:84% of the auditory system is frequency selectivity. 00:03:41.770 --> 00:03:43.370 align:middle line:84% So you saw this picture last time. 00:03:43.370 --> 00:03:46.890 align:middle line:84% These are tuning curves for auditory nerve fibers. 00:03:46.890 --> 00:03:50.330 align:middle line:84% So each one of these lines corresponds to a nerve fiber. 00:03:50.330 --> 00:03:55.450 align:middle line:84% And remember, each line plots the minimum sound intensity 00:03:55.450 --> 00:03:58.810 align:middle line:84% at a given frequency that is necessary to elicit 00:03:58.810 --> 00:04:00.630 align:middle line:90% a response in the fiber. 00:04:00.630 --> 00:04:02.470 align:middle line:84% So you hook an electrode into a nerve fiber. 00:04:02.470 --> 00:04:05.070 align:middle line:84% You're playing frequencies of different intensities. 00:04:05.070 --> 00:04:07.930 align:middle line:84% And you're looking at what the threshold of the nerve fiber is. 00:04:07.930 --> 00:04:10.612 align:middle line:84% And so for a particular nerve fiber, say this one, 00:04:10.612 --> 00:04:12.570 align:middle line:84% there is a special frequency, which we normally 00:04:12.570 --> 00:04:15.730 align:middle line:84% call the characteristic frequency, to which it responds 00:04:15.730 --> 00:04:16.970 align:middle line:90% best. 00:04:16.970 --> 00:04:18.950 align:middle line:84% So the amplitude of the sound doesn't 00:04:18.950 --> 00:04:20.950 align:middle line:84% have to be very high in order to get a response. 00:04:20.950 --> 00:04:23.630 align:middle line:84% And if you change the frequency, higher or lower, 00:04:23.630 --> 00:04:27.200 align:middle line:90% the threshold goes way up. 00:04:27.200 --> 00:04:31.250 align:middle line:84% So this empirical result suggests 00:04:31.250 --> 00:04:33.610 align:middle line:84% that the cochlea can be thought of as a bank 00:04:33.610 --> 00:04:36.490 align:middle line:90% of bandpass filters. 00:04:36.490 --> 00:04:38.570 align:middle line:84% And this property of frequency selectivity 00:04:38.570 --> 00:04:40.950 align:middle line:84% has a whole host of perceptual consequences. 00:04:40.950 --> 00:04:43.530 align:middle line:90% So this is just one. 00:04:43.530 --> 00:04:48.130 align:middle line:84% So how many of you are familiar with the phenomenon of beating? 00:04:48.130 --> 00:04:49.530 align:middle line:90% OK, a couple people. 00:04:49.530 --> 00:04:51.500 align:middle line:84% So if you play a string instrument 00:04:51.500 --> 00:04:53.340 align:middle line:84% you probably take advantage of beating 00:04:53.340 --> 00:04:55.420 align:middle line:90% when you tune your instrument. 00:04:55.420 --> 00:04:57.660 align:middle line:84% So beating is an acoustic phenomenon, 00:04:57.660 --> 00:05:00.780 align:middle line:84% and it happens anytime there are multiple frequencies that 00:05:00.780 --> 00:05:01.480 align:middle line:90% are present. 00:05:01.480 --> 00:05:04.740 align:middle line:84% And what's shown on the slide is an example of that. 00:05:04.740 --> 00:05:09.072 align:middle line:84% So the slide shows you the waveforms of two pure tones. 00:05:09.072 --> 00:05:10.780 align:middle line:84% Remember, a pure tone is a tone that just 00:05:10.780 --> 00:05:12.080 align:middle line:90% consists of a single frequency. 00:05:12.080 --> 00:05:13.080 align:middle line:90% It's a sinusoid. 00:05:13.080 --> 00:05:17.360 align:middle line:84% So we've got the red pure tone that's at one frequency. 00:05:17.360 --> 00:05:20.860 align:middle line:90% And that sounds like this. 00:05:20.860 --> 00:05:22.270 align:middle line:90% [PURE TONE] 00:05:22.270 --> 00:05:23.940 align:middle line:90% Can everybody hear that? 00:05:23.940 --> 00:05:24.940 align:middle line:90% OK. 00:05:24.940 --> 00:05:27.280 align:middle line:84% And we've got the blue pure tone, 00:05:27.280 --> 00:05:29.100 align:middle line:90% which is a different frequency. 00:05:29.100 --> 00:05:30.540 align:middle line:90% [PURE TONE] 00:05:30.540 --> 00:05:33.670 align:middle line:90% It's a little bit higher. 00:05:33.670 --> 00:05:38.460 align:middle line:84% So you can see that, because those two pure tones are 00:05:38.460 --> 00:05:41.020 align:middle line:84% different frequencies, there's this point in time 00:05:41.020 --> 00:05:42.938 align:middle line:90% here where the phases align. 00:05:42.938 --> 00:05:44.980 align:middle line:84% So the peak of the red curve aligns with the peak 00:05:44.980 --> 00:05:46.020 align:middle line:90% of the blue curve. 00:05:46.020 --> 00:05:48.520 align:middle line:84% And over time, they drift out of phase. 00:05:48.520 --> 00:05:50.750 align:middle line:84% And so you get to this moment in time, 00:05:50.750 --> 00:05:53.870 align:middle line:84% where the peak of the red curve aligns with the trough 00:05:53.870 --> 00:05:56.070 align:middle line:90% of the blue curve. 00:05:56.070 --> 00:05:59.590 align:middle line:84% Now, if you play those two pure tones at the same time, 00:05:59.590 --> 00:06:02.450 align:middle line:90% the waveforms add in the air. 00:06:02.450 --> 00:06:04.270 align:middle line:84% And so the sound that would actually 00:06:04.270 --> 00:06:05.790 align:middle line:84% occur if both of those frequencies 00:06:05.790 --> 00:06:07.950 align:middle line:84% were played at the same time is the black curve, 00:06:07.950 --> 00:06:11.510 align:middle line:84% which is just the sum of the red curve and the blue curve. 00:06:11.510 --> 00:06:14.110 align:middle line:84% And because of this particular relationship 00:06:14.110 --> 00:06:15.870 align:middle line:84% over time between the phases, you 00:06:15.870 --> 00:06:19.050 align:middle line:84% can see that the amplitude of the black curve waxes and wanes. 00:06:19.050 --> 00:06:20.160 align:middle line:90% So it's high here. 00:06:20.160 --> 00:06:21.910 align:middle line:84% And then by the time you get to this point 00:06:21.910 --> 00:06:25.070 align:middle line:84% in time, the red and the blue frequency, they cancel out. 00:06:25.070 --> 00:06:29.590 align:middle line:84% So that's called beating, this waxing and waning in amplitude. 00:06:29.590 --> 00:06:31.970 align:middle line:84% And I mentioned that if you play a string instrument, 00:06:31.970 --> 00:06:34.830 align:middle line:84% you are probably familiar with that because you actually 00:06:34.830 --> 00:06:36.300 align:middle line:84% often listen to beating if you're 00:06:36.300 --> 00:06:38.550 align:middle line:84% trying to tune your instrument, because it'll tell you 00:06:38.550 --> 00:06:40.993 align:middle line:84% whether the notes that are played on two strings, 00:06:40.993 --> 00:06:42.410 align:middle line:84% if you play them at the same time, 00:06:42.410 --> 00:06:45.830 align:middle line:84% are exactly the same or a little bit different. 00:06:45.830 --> 00:06:49.216 align:middle line:84% So this is what it sounds like in this particular case. 00:06:49.216 --> 00:06:50.847 align:middle line:90% [PURE TONES] 00:06:50.847 --> 00:06:53.180 align:middle line:84% So it sounds a little bit like a telephone or something. 00:06:53.180 --> 00:06:56.658 align:middle line:84% So that modulation is perceptually 00:06:56.658 --> 00:06:57.700 align:middle line:90% the corollary to beating. 00:06:57.700 --> 00:07:01.120 align:middle line:84% And the word that we typically use to refer to that sound 00:07:01.120 --> 00:07:02.880 align:middle line:90% is roughness. 00:07:02.880 --> 00:07:08.840 align:middle line:84% So we call a sound rough when it contains amplitude modulation. 00:07:08.840 --> 00:07:12.700 align:middle line:84% So beating is an acoustic phenomenon. 00:07:12.700 --> 00:07:16.560 align:middle line:84% It happens any time you have two different frequencies that are 00:07:16.560 --> 00:07:19.900 align:middle line:90% being played at the same time. 00:07:19.900 --> 00:07:22.800 align:middle line:84% However, you only hear the beating 00:07:22.800 --> 00:07:25.900 align:middle line:84% as roughness under particular conditions. 00:07:25.900 --> 00:07:29.660 align:middle line:84% And they turn out to be constrained by the cochlea. 00:07:29.660 --> 00:07:33.480 align:middle line:84% So the beats are only heard if the two frequency 00:07:33.480 --> 00:07:36.920 align:middle line:84% components fall within the bandwidth 00:07:36.920 --> 00:07:39.240 align:middle line:90% of the filters in your cochlea. 00:07:39.240 --> 00:07:41.520 align:middle line:84% And they also have to be in the same ear. 00:07:41.520 --> 00:07:43.820 align:middle line:84% So if we take the red frequency-- 00:07:43.820 --> 00:07:46.730 align:middle line:84% if we use headphones, we take the red frequency 00:07:46.730 --> 00:07:49.310 align:middle line:84% and we apply that to your right ear and the blue frequency 00:07:49.310 --> 00:07:51.090 align:middle line:84% and we apply that to your left ear, 00:07:51.090 --> 00:07:52.890 align:middle line:90% you will not hear the beats. 00:07:52.890 --> 00:07:56.450 align:middle line:84% So you only hear them if they enter the same ear 00:07:56.450 --> 00:07:58.330 align:middle line:84% and if the frequencies are sufficiently 00:07:58.330 --> 00:08:02.390 align:middle line:84% close that they stimulate the same auditory filter. 00:08:02.390 --> 00:08:04.650 align:middle line:84% So here's just a quick demonstration of that. 00:08:04.650 --> 00:08:08.150 align:middle line:84% So I'm going to play you three sounds. 00:08:08.150 --> 00:08:10.810 align:middle line:84% They each consist of two frequencies. 00:08:10.810 --> 00:08:12.450 align:middle line:84% But we're going to vary the difference 00:08:12.450 --> 00:08:13.630 align:middle line:90% between the frequencies. 00:08:13.630 --> 00:08:14.130 align:middle line:90% All right. 00:08:14.130 --> 00:08:16.150 align:middle line:84% So here, you have a one semitone difference. 00:08:16.150 --> 00:08:17.630 align:middle line:90% That's a 12th of an octave. 00:08:17.630 --> 00:08:19.505 align:middle line:84% So that's like the difference between a white 00:08:19.505 --> 00:08:21.130 align:middle line:90% and a black key on the piano. 00:08:21.130 --> 00:08:23.130 align:middle line:90% [ONE SEMITONE] 00:08:23.130 --> 00:08:25.250 align:middle line:90% And you can hear that. 00:08:25.250 --> 00:08:27.570 align:middle line:84% Here, it goes to three semitones. 00:08:27.570 --> 00:08:29.130 align:middle line:90% [THREE SEMITONES] 00:08:29.130 --> 00:08:31.090 align:middle line:84% And the beating becomes a lot more subtle. 00:08:31.090 --> 00:08:32.970 align:middle line:90% And here, it's out at eight. 00:08:32.970 --> 00:08:35.610 align:middle line:90% [EIGHT SEMITONES] 00:08:35.610 --> 00:08:38.770 align:middle line:84% And so at eight semitones, you can probably 00:08:38.770 --> 00:08:41.429 align:middle line:84% hear those two frequencies really as distinct things. 00:08:41.429 --> 00:08:43.215 align:middle line:84% You can almost attend to one or the other. 00:08:43.215 --> 00:08:44.840 align:middle line:84% And you don't hear the beating anymore. 00:08:44.840 --> 00:08:46.280 align:middle line:84% And it's not because the beating is not there. 00:08:46.280 --> 00:08:47.447 align:middle line:90% It's out there in the world. 00:08:47.447 --> 00:08:48.980 align:middle line:90% It's in the air. 00:08:48.980 --> 00:08:50.740 align:middle line:84% It just doesn't make it through your ear 00:08:50.740 --> 00:08:53.020 align:middle line:84% because those two frequencies get processed 00:08:53.020 --> 00:08:54.980 align:middle line:84% by different parts of the cochlea. 00:08:54.980 --> 00:08:58.060 align:middle line:84% So this is one respect in which the frequency 00:08:58.060 --> 00:09:00.880 align:middle line:84% selectivity of the cochlea really constrains what you hear. 00:09:00.880 --> 00:09:04.340 align:middle line:90% 00:09:04.340 --> 00:09:04.980 align:middle line:90% OK. 00:09:04.980 --> 00:09:07.020 align:middle line:84% So one of the most important attributes 00:09:07.020 --> 00:09:09.540 align:middle line:84% of the auditory system is frequency selectivity. 00:09:09.540 --> 00:09:11.593 align:middle line:84% The tuning that we observe in the auditory nerve 00:09:11.593 --> 00:09:13.260 align:middle line:84% suggests that the cochlea can be thought 00:09:13.260 --> 00:09:16.060 align:middle line:84% of as a bank of bandpass filters. 00:09:16.060 --> 00:09:18.420 align:middle line:84% But the problem is that, in humans, we can't 00:09:18.420 --> 00:09:19.680 align:middle line:90% measure the filters directly. 00:09:19.680 --> 00:09:22.540 align:middle line:84% So this picture that I just popped up, 00:09:22.540 --> 00:09:26.380 align:middle line:84% this is probably from a cat, or a rat, or a mouse, 00:09:26.380 --> 00:09:30.600 align:middle line:84% or some critter in which you can do an invasive experiment. 00:09:30.600 --> 00:09:33.380 align:middle line:84% And we don't have the ability to make these kinds of measurements 00:09:33.380 --> 00:09:36.020 align:middle line:84% directly in humans, because the auditory nerve is very, very 00:09:36.020 --> 00:09:39.460 align:middle line:90% difficult to get to. 00:09:39.460 --> 00:09:42.150 align:middle line:84% So in humans, if you want to understand the frequency 00:09:42.150 --> 00:09:43.830 align:middle line:84% selectivity in the cochlea, you have 00:09:43.830 --> 00:09:48.750 align:middle line:84% to infer it indirectly from some other type of experiment. 00:09:48.750 --> 00:09:51.630 align:middle line:84% And the principal way that human frequency selectivity 00:09:51.630 --> 00:09:56.390 align:middle line:84% has been explored is by making use of what's called masking. 00:09:56.390 --> 00:09:57.790 align:middle line:90% So what is masking? 00:09:57.790 --> 00:10:00.470 align:middle line:84% Well, masking is defined as the process 00:10:00.470 --> 00:10:04.110 align:middle line:84% by which the threshold of audibility for one sound 00:10:04.110 --> 00:10:08.350 align:middle line:84% is raised by the presence of another masking sound. 00:10:08.350 --> 00:10:11.470 align:middle line:84% So let's suppose there's a particular sound that you're 00:10:11.470 --> 00:10:12.870 align:middle line:90% supposed to be detecting. 00:10:12.870 --> 00:10:15.130 align:middle line:90% And you can hear it just fine. 00:10:15.130 --> 00:10:17.610 align:middle line:84% And then we play another sound at the same time. 00:10:17.610 --> 00:10:20.550 align:middle line:84% Under some conditions, that second sound 00:10:20.550 --> 00:10:23.690 align:middle line:84% will make it much harder to hear the first sound. 00:10:23.690 --> 00:10:25.750 align:middle line:90% And that's called masking. 00:10:25.750 --> 00:10:28.970 align:middle line:84% And this happens all the time in everyday life. 00:10:28.970 --> 00:10:32.330 align:middle line:84% And the reason for that is that sounds add in the air. 00:10:32.330 --> 00:10:34.630 align:middle line:84% And so any time there's more than one thing that's 00:10:34.630 --> 00:10:36.510 align:middle line:84% making sound, one of them is probably 00:10:36.510 --> 00:10:38.470 align:middle line:84% masking at least part of the other. 00:10:38.470 --> 00:10:41.030 align:middle line:84% So it's a very, very common phenomenon. 00:10:41.030 --> 00:10:44.850 align:middle line:90% 00:10:44.850 --> 00:10:49.170 align:middle line:84% And it turns out that masking is very closely related 00:10:49.170 --> 00:10:51.090 align:middle line:90% to frequency selectivity. 00:10:51.090 --> 00:10:53.450 align:middle line:84% So this is a graph, an old graph, 00:10:53.450 --> 00:10:57.675 align:middle line:84% that shows the masking of a target pure tone-- 00:10:57.675 --> 00:10:59.050 align:middle line:84% so it's a little beep that you're 00:10:59.050 --> 00:11:02.330 align:middle line:84% supposed to detect-- by either a narrowband noise 00:11:02.330 --> 00:11:05.770 align:middle line:90% masker or another pure tone. 00:11:05.770 --> 00:11:09.370 align:middle line:84% And so what the graph is plotting is the amount 00:11:09.370 --> 00:11:12.410 align:middle line:84% of masking-- and I'll tell you in a second what that means-- 00:11:12.410 --> 00:11:15.930 align:middle line:84% as a function of the frequency of the target tone. 00:11:15.930 --> 00:11:20.410 align:middle line:84% And there is a masker that is present 00:11:20.410 --> 00:11:22.290 align:middle line:90% at this particular frequency. 00:11:22.290 --> 00:11:26.250 align:middle line:84% And it can either be a noise band or a tone. 00:11:26.250 --> 00:11:29.330 align:middle line:84% And so the caption of the figure describes this in detail. 00:11:29.330 --> 00:11:33.370 align:middle line:84% So the pure tone is at 400 cycles per second, 400 hertz. 00:11:33.370 --> 00:11:36.670 align:middle line:84% And the narrow band of noise is centered at 410 hertz. 00:11:36.670 --> 00:11:39.740 align:middle line:90% So they're in this range here. 00:11:39.740 --> 00:11:42.180 align:middle line:84% Now, the amount of masking, that's 00:11:42.180 --> 00:11:46.180 align:middle line:84% the amount by which you have to increase the level of the target 00:11:46.180 --> 00:11:49.820 align:middle line:84% tone in order to be able to hear it when the masker is also 00:11:49.820 --> 00:11:51.295 align:middle line:90% present. 00:11:51.295 --> 00:11:52.920 align:middle line:84% All right, what does the graph tell us? 00:11:52.920 --> 00:11:55.660 align:middle line:84% Well, it tells us, when the target tone has 00:11:55.660 --> 00:11:59.380 align:middle line:84% a frequency that's either much lower or much higher 00:11:59.380 --> 00:12:03.605 align:middle line:84% than the masker, well, then the amount of masking is very small. 00:12:03.605 --> 00:12:05.980 align:middle line:84% You really don't have to increase the level of the target 00:12:05.980 --> 00:12:08.780 align:middle line:84% very much in order to be able to tell that it's there. 00:12:08.780 --> 00:12:13.080 align:middle line:84% But when the target is closer in frequency to the masker, 00:12:13.080 --> 00:12:16.320 align:middle line:84% you've got to crank up its level fairly significantly. 00:12:16.320 --> 00:12:18.900 align:middle line:84% And in particular, when it's very similar to the frequency 00:12:18.900 --> 00:12:21.300 align:middle line:84% of the masker, you might have to elevate its level 00:12:21.300 --> 00:12:22.600 align:middle line:90% by like 60 decibels. 00:12:22.600 --> 00:12:26.580 align:middle line:90% That's highly significant 00:12:26.580 --> 00:12:29.060 align:middle line:84% So in general, the amount of masking 00:12:29.060 --> 00:12:31.940 align:middle line:84% is determined by the similarity of the frequency 00:12:31.940 --> 00:12:33.040 align:middle line:90% between the signals. 00:12:33.040 --> 00:12:37.670 align:middle line:84% Now, you might have noticed that the curve for the pure tone 00:12:37.670 --> 00:12:40.590 align:middle line:84% masker has got all these funny bumps and dips. 00:12:40.590 --> 00:12:42.670 align:middle line:84% And one of the reasons for this is 00:12:42.670 --> 00:12:44.135 align:middle line:84% that, when you have two tones that 00:12:44.135 --> 00:12:45.510 align:middle line:84% are present at the same time, you 00:12:45.510 --> 00:12:47.910 align:middle line:90% get this phenomenon of beating. 00:12:47.910 --> 00:12:50.150 align:middle line:84% So if you're a participant in this experiment, 00:12:50.150 --> 00:12:52.170 align:middle line:84% trying to detect whether the tone is there, 00:12:52.170 --> 00:12:55.770 align:middle line:84% you can actually use the beating to help you detect the tone. 00:12:55.770 --> 00:12:58.470 align:middle line:84% So there are certain places where the tone gets 00:12:58.470 --> 00:12:59.730 align:middle line:90% a little bit easier to detect. 00:12:59.730 --> 00:13:04.150 align:middle line:84% So that's why you get these bumps and valleys in the graph, 00:13:04.150 --> 00:13:06.370 align:middle line:84% whereas, for the noise masker, you can't do that. 00:13:06.370 --> 00:13:09.030 align:middle line:84% And so you get a much smoother curve. 00:13:09.030 --> 00:13:11.230 align:middle line:84% But the key concept to take away from this 00:13:11.230 --> 00:13:14.370 align:middle line:84% is that masking is determined by frequency selectivity. 00:13:14.370 --> 00:13:18.470 align:middle line:84% And so we can use it in order to infer the shape of the filters 00:13:18.470 --> 00:13:22.190 align:middle line:90% that you have in your ears. 00:13:22.190 --> 00:13:22.690 align:middle line:90% All right. 00:13:22.690 --> 00:13:25.630 align:middle line:84% And so the general model by which we do this 00:13:25.630 --> 00:13:29.510 align:middle line:84% is what's called the power spectrum model of masking. 00:13:29.510 --> 00:13:34.120 align:middle line:84% So this considers the cochlea to be a bank of filters. 00:13:34.120 --> 00:13:37.440 align:middle line:84% And you measure masking by asking people 00:13:37.440 --> 00:13:41.200 align:middle line:84% to detect a tone in the presence of a noise masker. 00:13:41.200 --> 00:13:44.660 align:middle line:84% And so the assumption is that a signal is heard-- 00:13:44.660 --> 00:13:46.440 align:middle line:90% so a signal here is the tone. 00:13:46.440 --> 00:13:48.040 align:middle line:84% It's heard when the signal to noise 00:13:48.040 --> 00:13:51.120 align:middle line:84% ratio at the output of the filter that you're trying 00:13:51.120 --> 00:13:55.360 align:middle line:90% to measure exceeds some value. 00:13:55.360 --> 00:13:56.780 align:middle line:90% You're playing noise. 00:13:56.780 --> 00:13:58.960 align:middle line:84% And then you're playing the signal. 00:13:58.960 --> 00:14:02.000 align:middle line:84% And the idea is that some amount of the noise 00:14:02.000 --> 00:14:04.200 align:middle line:90% gets through the filter. 00:14:04.200 --> 00:14:06.840 align:middle line:84% And your ability to detect the tone 00:14:06.840 --> 00:14:10.320 align:middle line:84% will depend on the ratio between the power of the tone 00:14:10.320 --> 00:14:13.760 align:middle line:84% and the power of the noise heard through the filter 00:14:13.760 --> 00:14:18.080 align:middle line:84% that you're using to monitor the presence of the tone. 00:14:18.080 --> 00:14:18.580 align:middle line:90% All right. 00:14:18.580 --> 00:14:20.330 align:middle line:84% So we assume that there is a filter that's 00:14:20.330 --> 00:14:22.560 align:middle line:90% characterized by a shape. 00:14:22.560 --> 00:14:25.000 align:middle line:84% The masker is a noise signal that 00:14:25.000 --> 00:14:28.060 align:middle line:84% will be characterized by some spectrum 00:14:28.060 --> 00:14:30.100 align:middle line:84% So this is power as a function of frequency. 00:14:30.100 --> 00:14:31.850 align:middle line:84% And the filter's filters got some response 00:14:31.850 --> 00:14:33.410 align:middle line:90% as a function of frequency. 00:14:33.410 --> 00:14:36.170 align:middle line:84% And so the power of the noise signal 00:14:36.170 --> 00:14:39.090 align:middle line:84% that gets through the filter is given by this integral. 00:14:39.090 --> 00:14:41.490 align:middle line:84% It's essentially the dot product between the spectrum 00:14:41.490 --> 00:14:45.290 align:middle line:84% of the noise and the spectrum of the filter. 00:14:45.290 --> 00:14:47.450 align:middle line:84% And so the model posits that your ability 00:14:47.450 --> 00:14:51.010 align:middle line:84% to detect the signal will be a function of how much power 00:14:51.010 --> 00:14:52.410 align:middle line:90% gets through the filter. 00:14:52.410 --> 00:14:55.930 align:middle line:84% So let's just look at an example to build up some intuition here. 00:14:55.930 --> 00:15:00.090 align:middle line:84% So the idea is that there's this red thing in your ear. 00:15:00.090 --> 00:15:00.910 align:middle line:90% It's a filter. 00:15:00.910 --> 00:15:04.960 align:middle line:84% And we want to infer its properties. 00:15:04.960 --> 00:15:07.650 align:middle line:84% The noise signal is shown in gray. 00:15:07.650 --> 00:15:09.110 align:middle line:90% So this is frequency. 00:15:09.110 --> 00:15:11.050 align:middle line:84% So we're looking in the frequency domain. 00:15:11.050 --> 00:15:14.970 align:middle line:84% And this is plotting the masker power or the filter response 00:15:14.970 --> 00:15:16.430 align:middle line:90% as a function of frequency. 00:15:16.430 --> 00:15:17.750 align:middle line:90% So we've got a broadband noise. 00:15:17.750 --> 00:15:21.630 align:middle line:84% It's got all these frequencies in it. 00:15:21.630 --> 00:15:26.610 align:middle line:84% So the amount of noise power that gets through the filter 00:15:26.610 --> 00:15:29.600 align:middle line:84% will clearly be influenced by the level of the masker. 00:15:29.600 --> 00:15:32.600 align:middle line:84% So if we just increase the intensity of the masker, 00:15:32.600 --> 00:15:35.300 align:middle line:90% the noise power will increase. 00:15:35.300 --> 00:15:38.620 align:middle line:84% But it will also be dependent on the spectral properties 00:15:38.620 --> 00:15:39.660 align:middle line:90% of the masker. 00:15:39.660 --> 00:15:41.820 align:middle line:84% So here, we have a masker that's very narrowband. 00:15:41.820 --> 00:15:44.400 align:middle line:84% It's only got this narrow band of frequencies. 00:15:44.400 --> 00:15:46.660 align:middle line:84% And if we add low and high frequencies 00:15:46.660 --> 00:15:48.780 align:middle line:84% to make it more broadband, you can 00:15:48.780 --> 00:15:52.620 align:middle line:84% see that there's going to be more frequencies here 00:15:52.620 --> 00:15:53.940 align:middle line:90% that are hitting the filter. 00:15:53.940 --> 00:15:57.200 align:middle line:84% So the dot product of the filter and the noise will go up. 00:15:57.200 --> 00:16:01.082 align:middle line:90% 00:16:01.082 --> 00:16:05.260 align:middle line:84% Now, the noise power will also be affected by the filter shape. 00:16:05.260 --> 00:16:08.060 align:middle line:84% So of course, we're assuming that there's a fixed filter 00:16:08.060 --> 00:16:09.280 align:middle line:90% shape in your ear. 00:16:09.280 --> 00:16:10.920 align:middle line:90% But we don't know what it is. 00:16:10.920 --> 00:16:14.160 align:middle line:84% And so one hypothesis might be that it would have this shape. 00:16:14.160 --> 00:16:18.160 align:middle line:84% Another hypothesis might be that it would have this shape. 00:16:18.160 --> 00:16:20.980 align:middle line:84% If the filter is wider, there will be more noise power 00:16:20.980 --> 00:16:22.280 align:middle line:90% coming through it. 00:16:22.280 --> 00:16:25.160 align:middle line:84% And thus, it should make it harder to detect the tone. 00:16:25.160 --> 00:16:28.430 align:middle line:90% 00:16:28.430 --> 00:16:29.190 align:middle line:90% All right. 00:16:29.190 --> 00:16:32.790 align:middle line:84% So the general model that we're using here, in order 00:16:32.790 --> 00:16:35.790 align:middle line:84% to try to understand frequency selectivity via masking, 00:16:35.790 --> 00:16:37.450 align:middle line:84% makes a whole bunch of assumptions. 00:16:37.450 --> 00:16:39.670 align:middle line:90% It assumes the filter is linear. 00:16:39.670 --> 00:16:43.390 align:middle line:84% It assumes that, when you are performing this detection 00:16:43.390 --> 00:16:46.550 align:middle line:84% task of saying whether you hear the noise or not, 00:16:46.550 --> 00:16:50.630 align:middle line:84% you are only using one filter that's centered at the signal 00:16:50.630 --> 00:16:51.453 align:middle line:90% frequency. 00:16:51.453 --> 00:16:52.370 align:middle line:90% So this is the signal. 00:16:52.370 --> 00:16:53.010 align:middle line:90% It's a pure tone. 00:16:53.010 --> 00:16:54.385 align:middle line:84% Remember, it's got one frequency. 00:16:54.385 --> 00:16:56.470 align:middle line:84% That's why there's a spike there. 00:16:56.470 --> 00:16:58.910 align:middle line:84% And in reality, your ear consists 00:16:58.910 --> 00:17:01.630 align:middle line:84% of this big bank of filters because the cochlea 00:17:01.630 --> 00:17:03.069 align:middle line:90% covers all frequencies. 00:17:03.069 --> 00:17:05.069 align:middle line:84% But we assume that you're just using one of them 00:17:05.069 --> 00:17:07.829 align:middle line:90% in order to detect the tone. 00:17:07.829 --> 00:17:09.630 align:middle line:84% And we also assume that the detection 00:17:09.630 --> 00:17:12.690 align:middle line:84% is based entirely on the overall power at that filter's output. 00:17:12.690 --> 00:17:16.150 align:middle line:84% And so none of these assumptions are strictly true. 00:17:16.150 --> 00:17:17.990 align:middle line:84% But they often provide what we think 00:17:17.990 --> 00:17:22.270 align:middle line:84% is a fairly reasonable first approximation. 00:17:22.270 --> 00:17:22.770 align:middle line:90% All right. 00:17:22.770 --> 00:17:24.550 align:middle line:84% So the general method that would be 00:17:24.550 --> 00:17:27.880 align:middle line:84% used for measuring frequency selectivity in a person 00:17:27.880 --> 00:17:30.160 align:middle line:84% is to have some fixed signal, and then 00:17:30.160 --> 00:17:32.560 align:middle line:84% to manipulate the characteristics of the masker 00:17:32.560 --> 00:17:35.000 align:middle line:84% and to measure how the signal detection 00:17:35.000 --> 00:17:39.060 align:middle line:84% threshold will vary as the masker characteristics vary. 00:17:39.060 --> 00:17:41.240 align:middle line:84% And then we're going to try to deduce 00:17:41.240 --> 00:17:43.920 align:middle line:84% the characteristics of the auditory filter in your ear 00:17:43.920 --> 00:17:46.003 align:middle line:90% from the data. 00:17:46.003 --> 00:17:47.420 align:middle line:84% So let me just give you an example 00:17:47.420 --> 00:17:49.460 align:middle line:84% I think will make this pretty clear. 00:17:49.460 --> 00:17:53.040 align:middle line:84% So this is like the classic experiment 00:17:53.040 --> 00:17:56.560 align:middle line:84% that was really the first demonstration of this approach. 00:17:56.560 --> 00:18:01.320 align:middle line:84% And it's attributed to Fletcher, who was a scientist who worked 00:18:01.320 --> 00:18:05.160 align:middle line:90% at Bell Labs back in the 1940s. 00:18:05.160 --> 00:18:07.600 align:middle line:84% And so people that were at phone companies 00:18:07.600 --> 00:18:09.920 align:middle line:84% were very interested in understanding 00:18:09.920 --> 00:18:12.240 align:middle line:84% frequency selectivity in the human ear 00:18:12.240 --> 00:18:15.600 align:middle line:84% and also masking, in part because masking really 00:18:15.600 --> 00:18:18.800 align:middle line:84% determines the extent to which you can compress an audio signal 00:18:18.800 --> 00:18:21.800 align:middle line:84% and have it still sound OK to a person. 00:18:21.800 --> 00:18:25.183 align:middle line:84% And the essential concept is that, if you've 00:18:25.183 --> 00:18:27.600 align:middle line:84% got some sound with all these different frequencies in it, 00:18:27.600 --> 00:18:29.820 align:middle line:84% but you know that a whole bunch of the frequencies 00:18:29.820 --> 00:18:31.920 align:middle line:84% will be inaudible because of masking, 00:18:31.920 --> 00:18:33.540 align:middle line:90% you don't have to encode them. 00:18:33.540 --> 00:18:37.240 align:middle line:84% And so you can save bits in the signal that you're sending, 00:18:37.240 --> 00:18:38.640 align:middle line:90% say over a phone line. 00:18:38.640 --> 00:18:42.540 align:middle line:84% And so nowadays, all audio compression algorithms 00:18:42.540 --> 00:18:46.880 align:middle line:84% essentially have built into them measurements of human masking. 00:18:46.880 --> 00:18:50.400 align:middle line:84% So MP3 codecs have that built into them. 00:18:50.400 --> 00:18:54.100 align:middle line:84% So people were interested in this for a long time. 00:18:54.100 --> 00:18:54.600 align:middle line:90% All right. 00:18:54.600 --> 00:18:57.940 align:middle line:84% So the experiment that Fletcher conducted 00:18:57.940 --> 00:19:01.260 align:middle line:84% involved measuring the signal detection threshold 00:19:01.260 --> 00:19:04.060 align:middle line:84% as a function of the noise bandwidth. 00:19:04.060 --> 00:19:07.400 align:middle line:84% So this is a noise that's got a constant spectral level. 00:19:07.400 --> 00:19:10.060 align:middle line:84% And we're just going to vary the bandwidth of the noise 00:19:10.060 --> 00:19:12.260 align:middle line:84% and measure how that affects your ability 00:19:12.260 --> 00:19:17.550 align:middle line:84% to detect a tone that's at the middle of the noise. 00:19:17.550 --> 00:19:21.890 align:middle line:84% And so the results of the experiment are shown over here. 00:19:21.890 --> 00:19:26.990 align:middle line:84% So this graph plots the signal threshold in dB SPL 00:19:26.990 --> 00:19:31.310 align:middle line:84% as a function of the masking noise bandwidth in hertz. 00:19:31.310 --> 00:19:37.790 align:middle line:84% So over here, the masker is very narrow and the threshold is low. 00:19:37.790 --> 00:19:40.430 align:middle line:84% As you increase the bandwidth of the masker, 00:19:40.430 --> 00:19:42.070 align:middle line:90% the threshold goes up. 00:19:42.070 --> 00:19:44.630 align:middle line:84% And then there's this magical point 00:19:44.630 --> 00:19:47.190 align:middle line:84% at which, if you continue to increase 00:19:47.190 --> 00:19:50.670 align:middle line:84% the bandwidth of the masker, the signal threshold does not 00:19:50.670 --> 00:19:52.450 align:middle line:90% continue to increase. 00:19:52.450 --> 00:19:55.730 align:middle line:84% And the intuition here is that, at this point, 00:19:55.730 --> 00:19:58.070 align:middle line:84% you've hit the bandwidth of the filter 00:19:58.070 --> 00:20:00.070 align:middle line:84% that you're using to detect the signal. 00:20:00.070 --> 00:20:02.150 align:middle line:84% So if you increase the bandwidth of the masker 00:20:02.150 --> 00:20:05.808 align:middle line:84% further, beyond the edges of the red curve, 00:20:05.808 --> 00:20:07.350 align:middle line:84% it's not going to affect your ability 00:20:07.350 --> 00:20:09.870 align:middle line:84% to detect the tone because those frequencies don't 00:20:09.870 --> 00:20:11.750 align:middle line:90% get passed through the filter. 00:20:11.750 --> 00:20:14.410 align:middle line:84% And that's called the critical band. 00:20:14.410 --> 00:20:17.262 align:middle line:90% 00:20:17.262 --> 00:20:19.720 align:middle line:84% So we're going to demonstrate this and measure 00:20:19.720 --> 00:20:23.040 align:middle line:90% our own critical bands. 00:20:23.040 --> 00:20:26.400 align:middle line:84% What we're going to do is an experiment 00:20:26.400 --> 00:20:28.600 align:middle line:84% where I will play you a sequence of tones 00:20:28.600 --> 00:20:30.640 align:middle line:90% that will decrease in level. 00:20:30.640 --> 00:20:33.400 align:middle line:84% And your job is to count how many of them you hear. 00:20:33.400 --> 00:20:36.440 align:middle line:84% And the idea is the number that you 00:20:36.440 --> 00:20:40.600 align:middle line:84% can hear will be proportional to your-- inversely proportional 00:20:40.600 --> 00:20:41.500 align:middle line:90% to your threshold. 00:20:41.500 --> 00:20:43.240 align:middle line:84% All right, so let's check it out. 00:20:43.240 --> 00:20:44.240 align:middle line:90% [AUDIO PLAYBACK] 00:20:44.240 --> 00:20:47.640 align:middle line:90% - Critical bands by masking. 00:20:47.640 --> 00:20:52.220 align:middle line:84% You will hear a 2,000 hertz tone in 10 decreasing steps of 5 00:20:52.220 --> 00:20:53.160 align:middle line:90% decibels. 00:20:53.160 --> 00:20:56.000 align:middle line:84% Count how many steps you can hear. 00:20:56.000 --> 00:20:59.744 align:middle line:90% Series are presented twice. 00:20:59.744 --> 00:21:04.624 align:middle line:90% [DECREASING TONES] 00:21:04.624 --> 00:21:09.992 align:middle line:90% 00:21:09.992 --> 00:21:10.968 align:middle line:90% [END PLAYBACK] 00:21:10.968 --> 00:21:12.460 align:middle line:84% JOSH MCDERMOTT: Did you all get 10? 00:21:12.460 --> 00:21:12.960 align:middle line:90% Yeah? 00:21:12.960 --> 00:21:13.960 align:middle line:90% All right, good. 00:21:13.960 --> 00:21:14.540 align:middle line:90% OK. 00:21:14.540 --> 00:21:15.620 align:middle line:90% So you can hear 10. 00:21:15.620 --> 00:21:17.370 align:middle line:84% Now, we're going to do the same thing. 00:21:17.370 --> 00:21:19.210 align:middle line:84% But there's going to be a noise masker that 00:21:19.210 --> 00:21:20.630 align:middle line:90% will be present throughout. 00:21:20.630 --> 00:21:23.170 align:middle line:84% And so the tones will get harder to hear. 00:21:23.170 --> 00:21:26.756 align:middle line:84% And you won't be able to hear quite as many. 00:21:26.756 --> 00:21:28.025 align:middle line:90% [AUDIO PLAYBACK] 00:21:28.025 --> 00:21:33.226 align:middle line:84% - Finally, the bandwidth is reduced to only 10 Hertz. 00:21:33.226 --> 00:21:44.938 align:middle line:90% [DECREASING TONES AND MASK] 00:21:44.938 --> 00:21:46.890 align:middle line:90% [END PLAYBACK] 00:21:46.890 --> 00:21:48.090 align:middle line:90% JOSH MCDERMOTT: I got eight. 00:21:48.090 --> 00:21:49.570 align:middle line:90% Is that about right? 00:21:49.570 --> 00:21:53.530 align:middle line:84% And I'm going to actually let you watch this. 00:21:53.530 --> 00:21:54.030 align:middle line:90% All right. 00:21:54.030 --> 00:21:56.550 align:middle line:84% So that was a very narrowband masker. 00:21:56.550 --> 00:21:58.070 align:middle line:90% But it elevated your threshold. 00:21:58.070 --> 00:22:00.410 align:middle line:90% So initially, you could hear 10. 00:22:00.410 --> 00:22:02.430 align:middle line:84% Now, after that, you could hear eight or nine. 00:22:02.430 --> 00:22:03.222 align:middle line:90% Is that consistent? 00:22:03.222 --> 00:22:04.530 align:middle line:90% Anybody disagree? 00:22:04.530 --> 00:22:05.490 align:middle line:90% No? 00:22:05.490 --> 00:22:07.610 align:middle line:90% All right. 00:22:07.610 --> 00:22:09.650 align:middle line:84% I hear some complaints from the crowd. 00:22:09.650 --> 00:22:12.023 align:middle line:84% All right, let's make it wider still. 00:22:12.023 --> 00:22:12.690 align:middle line:90% [AUDIO PLAYBACK] 00:22:12.690 --> 00:22:18.264 align:middle line:84% - Next, noise with a bandwidth of 250 hertz is used. 00:22:18.264 --> 00:22:28.912 align:middle line:90% [DECREASING TONES AND MASK] 00:22:28.912 --> 00:22:29.900 align:middle line:90% [END PLAYBACK] 00:22:29.900 --> 00:22:31.740 align:middle line:90% JOSH MCDERMOTT: I got five. 00:22:31.740 --> 00:22:32.380 align:middle line:90% Yeah? 00:22:32.380 --> 00:22:33.360 align:middle line:90% Is that about right? 00:22:33.360 --> 00:22:34.680 align:middle line:90% OK. 00:22:34.680 --> 00:22:37.120 align:middle line:84% All right, so the threshold was elevated, 00:22:37.120 --> 00:22:39.763 align:middle line:84% went from eight or nine down to five. 00:22:39.763 --> 00:22:41.180 align:middle line:84% What's going to happen here, we're 00:22:41.180 --> 00:22:42.733 align:middle line:90% going to keep making it wider. 00:22:42.733 --> 00:22:44.900 align:middle line:84% And at some point, it's going to stop getting harder 00:22:44.900 --> 00:22:46.940 align:middle line:90% to hear the sounds. 00:22:46.940 --> 00:22:48.060 align:middle line:90% [AUDIO PLAYBACK] 00:22:48.060 --> 00:22:54.460 align:middle line:84% - Next, the noise has a bandwidth of 1,000 hertz. 00:22:54.460 --> 00:23:05.684 align:middle line:90% [DECREASING TONES AND MASK] 00:23:05.684 --> 00:23:06.415 align:middle line:90% [END PLAYBACK] 00:23:06.415 --> 00:23:08.040 align:middle line:84% JOSH MCDERMOTT: I still got about five. 00:23:08.040 --> 00:23:08.440 align:middle line:90% All right? 00:23:08.440 --> 00:23:09.523 align:middle line:90% All right, so it's stable. 00:23:09.523 --> 00:23:12.380 align:middle line:84% Now, we're going to go to something that's a lot wider. 00:23:12.380 --> 00:23:13.047 align:middle line:90% [AUDIO PLAYBACK] 00:23:13.047 --> 00:23:16.030 align:middle line:84% - Now, the signal is masked with broadband noise. 00:23:16.030 --> 00:23:28.614 align:middle line:90% [DECREASING TONES AND BROADBAND] 00:23:28.614 --> 00:23:30.066 align:middle line:90% [END PLAYBACK] 00:23:30.066 --> 00:23:31.790 align:middle line:84% JOSH MCDERMOTT: And what's kind of 00:23:31.790 --> 00:23:34.270 align:middle line:84% cool about that is that last noise is 00:23:34.270 --> 00:23:38.310 align:middle line:84% way louder than the one that came before it. 00:23:38.310 --> 00:23:40.510 align:middle line:84% But it's no harder to detect the tone. 00:23:40.510 --> 00:23:43.030 align:middle line:84% And the explanation for that is that you're 00:23:43.030 --> 00:23:46.710 align:middle line:84% using the filters in your ear to do that detection. 00:23:46.710 --> 00:23:49.750 align:middle line:84% And the response of the filter is not affected very much 00:23:49.750 --> 00:23:51.230 align:middle line:84% by all those extra frequencies that 00:23:51.230 --> 00:23:53.330 align:middle line:90% get added above and below it. 00:23:53.330 --> 00:23:56.630 align:middle line:90% So that's the critical band. 00:23:56.630 --> 00:23:57.190 align:middle line:90% All right. 00:23:57.190 --> 00:24:02.130 align:middle line:84% Any questions about this masking experiment or the general logic? 00:24:02.130 --> 00:24:03.070 align:middle line:90% Yeah? 00:24:03.070 --> 00:24:05.430 align:middle line:84% STUDENT: Is it supposed to be that you hear 00:24:05.430 --> 00:24:07.730 align:middle line:90% more as it keeps getting wider? 00:24:07.730 --> 00:24:09.522 align:middle line:84% Because at least for me, I thought 00:24:09.522 --> 00:24:12.140 align:middle line:84% I heard more for the last one than the middle two. 00:24:12.140 --> 00:24:15.080 align:middle line:84% But I don't know if that's like an effect or not or just 00:24:15.080 --> 00:24:16.152 align:middle line:90% happens to be like that. 00:24:16.152 --> 00:24:17.860 align:middle line:84% JOSH MCDERMOTT: My guess is it's a fluke. 00:24:17.860 --> 00:24:20.760 align:middle line:84% I mean, there's no real reason to think that that would happen. 00:24:20.760 --> 00:24:26.320 align:middle line:90% 00:24:26.320 --> 00:24:29.040 align:middle line:84% The fact that you actually hear a tone versus not 00:24:29.040 --> 00:24:31.840 align:middle line:84% is somewhat subject to random effects. 00:24:31.840 --> 00:24:34.440 align:middle line:84% Sometimes, you're paying a little bit more attention, 00:24:34.440 --> 00:24:36.940 align:middle line:84% or had a little bit more coffee, or whatever. 00:24:36.940 --> 00:24:40.580 align:middle line:84% And so from trial to trial, there'll be fluctuations. 00:24:40.580 --> 00:24:42.162 align:middle line:84% But yeah, on average, I don't think 00:24:42.162 --> 00:24:44.120 align:middle line:84% we would expect that you'd be any better there. 00:24:44.120 --> 00:24:46.200 align:middle line:90% It typically flattens out, yeah. 00:24:46.200 --> 00:24:49.260 align:middle line:90% Any other questions? 00:24:49.260 --> 00:24:49.760 align:middle line:90% OK. 00:24:49.760 --> 00:24:56.120 align:middle line:90% 00:24:56.120 --> 00:24:58.780 align:middle line:84% The experiment here, it's conceptually very powerful. 00:24:58.780 --> 00:25:00.240 align:middle line:90% It makes a really good demo. 00:25:00.240 --> 00:25:03.160 align:middle line:84% And it's great for teaching purposes. 00:25:03.160 --> 00:25:05.640 align:middle line:84% But experimentally, in terms of actually 00:25:05.640 --> 00:25:08.623 align:middle line:84% using this to deduce the properties of the ear, 00:25:08.623 --> 00:25:10.040 align:middle line:84% it's actually not all that useful. 00:25:10.040 --> 00:25:12.060 align:middle line:90% And so nobody uses this anymore. 00:25:12.060 --> 00:25:16.100 align:middle line:84% And the main issue here is that the filter output power 00:25:16.100 --> 00:25:18.180 align:middle line:84% is really dominated by frequencies that are 00:25:18.180 --> 00:25:19.840 align:middle line:90% near the center of the filter. 00:25:19.840 --> 00:25:24.860 align:middle line:84% And so imagine you're comparing this versus this. 00:25:24.860 --> 00:25:27.460 align:middle line:84% And because the filter has a weak response 00:25:27.460 --> 00:25:29.955 align:middle line:84% here in the skirts, the frequencies out there 00:25:29.955 --> 00:25:31.580 align:middle line:84% actually don't have a huge contribution 00:25:31.580 --> 00:25:33.840 align:middle line:90% to the overall amount of power. 00:25:33.840 --> 00:25:35.340 align:middle line:84% And so the technique is actually not 00:25:35.340 --> 00:25:38.887 align:middle line:84% super useful for providing an accurate estimate of the filter 00:25:38.887 --> 00:25:39.720 align:middle line:90% shape and bandwidth. 00:25:39.720 --> 00:25:43.180 align:middle line:84% So we mostly just now use this for teaching purposes. 00:25:43.180 --> 00:25:47.220 align:middle line:84% So the standard modern method for getting estimates 00:25:47.220 --> 00:25:48.940 align:middle line:84% of frequency selectivity is what's 00:25:48.940 --> 00:25:51.180 align:middle line:90% called the notched-noise method. 00:25:51.180 --> 00:25:53.320 align:middle line:84% So the idea is what's shown here. 00:25:53.320 --> 00:25:55.280 align:middle line:84% So you're still trying to detect a signal. 00:25:55.280 --> 00:25:57.220 align:middle line:84% But now, instead of having noise that 00:25:57.220 --> 00:26:01.360 align:middle line:84% has this continuous spectrum, the noise has a notch in it. 00:26:01.360 --> 00:26:04.000 align:middle line:84% So you're supplying frequencies here and frequencies here. 00:26:04.000 --> 00:26:07.230 align:middle line:84% And then you can change the position of each 00:26:07.230 --> 00:26:08.910 align:middle line:90% of the bands of frequencies. 00:26:08.910 --> 00:26:12.810 align:middle line:84% So you can compare this, to say this, to say this. 00:26:12.810 --> 00:26:14.350 align:middle line:84% And so you measure your threshold 00:26:14.350 --> 00:26:16.050 align:middle line:84% for these different types of maskers. 00:26:16.050 --> 00:26:20.910 align:middle line:84% And so the idea is that, in the black case, 00:26:20.910 --> 00:26:23.480 align:middle line:84% the only noise power that's coming through the filter 00:26:23.480 --> 00:26:25.730 align:middle line:84% is really the things that are out there in the skirts. 00:26:25.730 --> 00:26:29.390 align:middle line:84% And so you isolate the response at that particular point 00:26:29.390 --> 00:26:30.170 align:middle line:90% in the filter. 00:26:30.170 --> 00:26:33.790 align:middle line:84% And you can better characterize it. 00:26:33.790 --> 00:26:34.290 align:middle line:90% All right. 00:26:34.290 --> 00:26:37.830 align:middle line:84% And so there's now been lots of experiments along these lines. 00:26:37.830 --> 00:26:39.547 align:middle line:84% I'm not going to work through in detail 00:26:39.547 --> 00:26:41.630 align:middle line:84% how you would go from the results of an experiment 00:26:41.630 --> 00:26:44.830 align:middle line:84% like this to inferring the exact filter shape. 00:26:44.830 --> 00:26:46.250 align:middle line:90% But you can work that out. 00:26:46.250 --> 00:26:49.870 align:middle line:84% And this is just an example of a derived filter function, where 00:26:49.870 --> 00:26:52.750 align:middle line:84% you have the response of this measured filter 00:26:52.750 --> 00:26:54.830 align:middle line:90% as a function of frequency. 00:26:54.830 --> 00:26:57.790 align:middle line:84% And so the notched-noise method is generally 00:26:57.790 --> 00:27:00.970 align:middle line:84% acknowledged to be better than Fletcher's band-widening method. 00:27:00.970 --> 00:27:03.870 align:middle line:84% So it gives you more information about the filter skirts. 00:27:03.870 --> 00:27:05.400 align:middle line:84% It's also got this advantage that it 00:27:05.400 --> 00:27:08.960 align:middle line:84% reduces the potential for what's called off-frequency listening. 00:27:08.960 --> 00:27:10.785 align:middle line:84% So when you're doing this experiment, 00:27:10.785 --> 00:27:12.160 align:middle line:84% the whole point of the experiment 00:27:12.160 --> 00:27:16.200 align:middle line:84% is to try to characterize the red filter. 00:27:16.200 --> 00:27:19.120 align:middle line:84% It's the one that's centered at the signal frequency. 00:27:19.120 --> 00:27:21.620 align:middle line:84% But the observer, the person who's doing the experiment, 00:27:21.620 --> 00:27:24.440 align:middle line:84% they have the option to use their entire ear. 00:27:24.440 --> 00:27:26.720 align:middle line:84% And in actuality, you have these other filters 00:27:26.720 --> 00:27:29.470 align:middle line:84% that are positioned at slightly different frequencies. 00:27:29.470 --> 00:27:31.720 align:middle line:84% And in principle, you might actually get better signal 00:27:31.720 --> 00:27:34.320 align:middle line:84% to noise ratios under some conditions listening 00:27:34.320 --> 00:27:35.860 align:middle line:90% to adjacent filters. 00:27:35.860 --> 00:27:37.760 align:middle line:84% But the idea of the notched-noise signal 00:27:37.760 --> 00:27:40.920 align:middle line:84% is that the filter that's actually centered at the signal 00:27:40.920 --> 00:27:43.360 align:middle line:84% frequency is the one where the signal to noise ratio 00:27:43.360 --> 00:27:44.960 align:middle line:90% would always be the best. 00:27:44.960 --> 00:27:47.800 align:middle line:84% And so the observer just doesn't have any reason 00:27:47.800 --> 00:27:49.300 align:middle line:90% to use anything other than that. 00:27:49.300 --> 00:27:52.960 align:middle line:84% So that's considered to be an advantage. 00:27:52.960 --> 00:27:53.460 align:middle line:90% All right. 00:27:53.460 --> 00:27:56.360 align:middle line:84% So this has now been used fairly extensively in people 00:27:56.360 --> 00:27:58.760 align:middle line:84% to characterize frequency selectivity. 00:27:58.760 --> 00:28:02.160 align:middle line:84% So one of the main findings that comes out of these experiments 00:28:02.160 --> 00:28:04.530 align:middle line:84% is that frequency selectivity varies 00:28:04.530 --> 00:28:06.510 align:middle line:84% as a function of the center frequency. 00:28:06.510 --> 00:28:11.190 align:middle line:84% So this is a graph that is plotting absolute bandwidths. 00:28:11.190 --> 00:28:14.333 align:middle line:84% And that's the equivalent rectangular bandwidth. 00:28:14.333 --> 00:28:16.250 align:middle line:84% So it's one measure of how wide the filter is, 00:28:16.250 --> 00:28:19.810 align:middle line:84% measured in hertz, as a function of the characteristic frequency. 00:28:19.810 --> 00:28:21.970 align:middle line:84% And so down here at 100 hertz, you 00:28:21.970 --> 00:28:25.150 align:middle line:84% can see that the bandwidth is like maybe 35 hertz. 00:28:25.150 --> 00:28:28.770 align:middle line:84% And then if you go up to a kilohertz, it's about 150 hertz. 00:28:28.770 --> 00:28:30.410 align:middle line:84% If you go up to 5 kilohertz, you can 00:28:30.410 --> 00:28:32.490 align:middle line:90% see that it's like 500 hertz. 00:28:32.490 --> 00:28:39.090 align:middle line:84% So as you move along the cochlea, the bandwidths vary. 00:28:39.090 --> 00:28:42.010 align:middle line:84% This is a pretty general property biological auditory 00:28:42.010 --> 00:28:44.090 align:middle line:84% systems, that the bandwidths tend 00:28:44.090 --> 00:28:48.970 align:middle line:84% to be higher at higher frequencies in absolute terms. 00:28:48.970 --> 00:28:52.510 align:middle line:84% Now, if you look at the relative bandwidths, 00:28:52.510 --> 00:28:55.410 align:middle line:84% so if you express the bandwidth relative to the center 00:28:55.410 --> 00:28:58.170 align:middle line:84% frequency, they actually stay pretty constant 00:28:58.170 --> 00:28:59.590 align:middle line:90% or even decrease. 00:28:59.590 --> 00:29:02.980 align:middle line:84% And so these are filters plotted on a logarithmic frequency 00:29:02.980 --> 00:29:03.780 align:middle line:90% scale. 00:29:03.780 --> 00:29:05.620 align:middle line:84% And you can now see that they look 00:29:05.620 --> 00:29:08.640 align:middle line:84% pretty similar at different points along the spectrum 00:29:08.640 --> 00:29:10.140 align:middle line:84% and, in fact, maybe are a little bit 00:29:10.140 --> 00:29:12.100 align:middle line:84% narrower here at high frequencies 00:29:12.100 --> 00:29:13.580 align:middle line:90% than at low frequencies. 00:29:13.580 --> 00:29:16.500 align:middle line:84% And again, this is a fairly general property 00:29:16.500 --> 00:29:18.040 align:middle line:90% biological auditory systems. 00:29:18.040 --> 00:29:20.540 align:middle line:84% The filters have that property that, when 00:29:20.540 --> 00:29:22.440 align:middle line:84% you look at them on a logarithmic scale, 00:29:22.440 --> 00:29:24.940 align:middle line:90% they looked pretty constant. 00:29:24.940 --> 00:29:28.980 align:middle line:84% So as I mentioned, there's lots of applications 00:29:28.980 --> 00:29:31.460 align:middle line:84% of these measurements of frequency selectivity. 00:29:31.460 --> 00:29:35.860 align:middle line:84% One of them has to do with using masking for audio compression. 00:29:35.860 --> 00:29:38.900 align:middle line:84% Another is that this is a pretty commonly used diagnostic 00:29:38.900 --> 00:29:41.660 align:middle line:84% to tell whether somebody's ears are healthy. 00:29:41.660 --> 00:29:44.180 align:middle line:84% So as we'll discuss, when you start to lose your hearing, 00:29:44.180 --> 00:29:47.540 align:middle line:90% the filters tend to get broader. 00:29:47.540 --> 00:29:49.500 align:middle line:84% And then another application actually 00:29:49.500 --> 00:29:52.740 align:middle line:84% involves comparing the hearing of people 00:29:52.740 --> 00:29:55.060 align:middle line:84% to the hearing of non-human animals. 00:29:55.060 --> 00:29:57.380 align:middle line:84% And that's interesting for a whole bunch of reasons. 00:29:57.380 --> 00:30:00.110 align:middle line:84% Oftentimes, we're stuck with performing experiments 00:30:00.110 --> 00:30:04.310 align:middle line:84% on non-human animals because they're invasive methods that 00:30:04.310 --> 00:30:05.588 align:middle line:90% are not possible in humans. 00:30:05.588 --> 00:30:07.630 align:middle line:84% It's also really interesting, from the standpoint 00:30:07.630 --> 00:30:09.470 align:middle line:84% of thinking about evolution, to understand 00:30:09.470 --> 00:30:12.550 align:middle line:84% the relationship between humans and non-human animals. 00:30:12.550 --> 00:30:16.030 align:middle line:84% And indeed, out of this type of work, 00:30:16.030 --> 00:30:19.190 align:middle line:84% there's emerged this interesting evidence 00:30:19.190 --> 00:30:22.070 align:middle line:84% that humans appear to have substantially better frequency 00:30:22.070 --> 00:30:24.390 align:middle line:84% selectivity than non-human animals. 00:30:24.390 --> 00:30:27.710 align:middle line:84% So it's like our ears are more tightly tuned. 00:30:27.710 --> 00:30:32.390 align:middle line:84% And so this is one example graph that shows an example of this. 00:30:32.390 --> 00:30:35.610 align:middle line:84% So in this literature, they tend to not plot bandwidths. 00:30:35.610 --> 00:30:38.050 align:middle line:84% Instead, they plot the tuning sharpness, 00:30:38.050 --> 00:30:41.350 align:middle line:84% which is just reciprocally related to the bandwidth. 00:30:41.350 --> 00:30:43.190 align:middle line:90% So this is the tuning sharpness. 00:30:43.190 --> 00:30:45.433 align:middle line:84% And this is the characteristic frequency. 00:30:45.433 --> 00:30:47.350 align:middle line:84% And there are measurements that were made here 00:30:47.350 --> 00:30:48.610 align:middle line:90% in cats and guinea pigs. 00:30:48.610 --> 00:30:50.110 align:middle line:90% That's red and blue. 00:30:50.110 --> 00:30:54.030 align:middle line:84% And then the black lines plot two different estimates 00:30:54.030 --> 00:30:55.770 align:middle line:90% of human tuning bandwidth. 00:30:55.770 --> 00:30:58.960 align:middle line:84% And the point is that the black lines are substantially 00:30:58.960 --> 00:31:00.500 align:middle line:84% above the blue and the red lines. 00:31:00.500 --> 00:31:04.860 align:middle line:84% So the filters in the human are thought to be more sharp. 00:31:04.860 --> 00:31:08.240 align:middle line:90% 00:31:08.240 --> 00:31:13.360 align:middle line:84% Here's another graph that shows cats, macaques, and humans. 00:31:13.360 --> 00:31:16.000 align:middle line:84% And here, you can see that the macaque monkey 00:31:16.000 --> 00:31:20.280 align:middle line:84% seems to be somewhere between cats and humans. 00:31:20.280 --> 00:31:23.812 align:middle line:84% And it's interesting to think about why this is. 00:31:23.812 --> 00:31:26.020 align:middle line:84% And really, there isn't widespread agreement on this. 00:31:26.020 --> 00:31:28.160 align:middle line:84% It's just an empirical observation. 00:31:28.160 --> 00:31:31.720 align:middle line:84% You might imagine that some of evolutionary pressure 00:31:31.720 --> 00:31:34.920 align:middle line:84% for communication might have driven this. 00:31:34.920 --> 00:31:36.720 align:middle line:84% Other people think it's just a function 00:31:36.720 --> 00:31:37.980 align:middle line:90% of the size of the cochlea. 00:31:37.980 --> 00:31:39.180 align:middle line:90% We don't really know. 00:31:39.180 --> 00:31:43.600 align:middle line:84% But it's an interesting difference. 00:31:43.600 --> 00:31:44.280 align:middle line:90% All right. 00:31:44.280 --> 00:31:47.840 align:middle line:84% Now, I told you a little bit about some of these properties 00:31:47.840 --> 00:31:51.200 align:middle line:84% that we find in biological auditory systems. 00:31:51.200 --> 00:31:56.520 align:middle line:84% And it's natural to wonder why the ear is the way that it is. 00:31:56.520 --> 00:31:59.340 align:middle line:90% 00:31:59.340 --> 00:32:02.980 align:middle line:84% And some recent approaches that are now possible 00:32:02.980 --> 00:32:05.300 align:middle line:84% have given us some new ways of trying 00:32:05.300 --> 00:32:08.420 align:middle line:84% to understand these questions, where advances in machine 00:32:08.420 --> 00:32:11.180 align:middle line:84% learning now make it possible to optimize 00:32:11.180 --> 00:32:15.420 align:middle line:84% machine systems to solve interesting perceptual tasks. 00:32:15.420 --> 00:32:19.940 align:middle line:84% And so this is a cool paper that came out of Google 00:32:19.940 --> 00:32:24.640 align:middle line:84% a few years ago now, where they, as the title says, 00:32:24.640 --> 00:32:29.460 align:middle line:84% learned the front-end of a neural network that 00:32:29.460 --> 00:32:30.880 align:middle line:90% had to recognize speech. 00:32:30.880 --> 00:32:35.660 align:middle line:90% 00:32:35.660 --> 00:32:37.413 align:middle line:84% I mean, Google was not really interested 00:32:37.413 --> 00:32:40.080 align:middle line:84% in the scientific questions that we're interested in this class. 00:32:40.080 --> 00:32:41.660 align:middle line:84% They were just trying to build a better speech recognition 00:32:41.660 --> 00:32:42.460 align:middle line:90% system. 00:32:42.460 --> 00:32:45.780 align:middle line:84% But the result of the study actually 00:32:45.780 --> 00:32:47.700 align:middle line:84% ends up being kind of informative with respect 00:32:47.700 --> 00:32:49.360 align:middle line:84% to thinking about the auditory system. 00:32:49.360 --> 00:32:54.270 align:middle line:84% So this is a big typical system for recognizing speech. 00:32:54.270 --> 00:32:57.030 align:middle line:84% So you start out with a sound waveform. 00:32:57.030 --> 00:33:01.630 align:middle line:84% And in this case, the system convolves the waveform 00:33:01.630 --> 00:33:03.350 align:middle line:90% with a bunch of filters. 00:33:03.350 --> 00:33:05.990 align:middle line:84% And these filters, critically, were learnable. 00:33:05.990 --> 00:33:09.205 align:middle line:84% So there's convolution with these filters. 00:33:09.205 --> 00:33:10.830 align:middle line:84% There's some pretty standard operations 00:33:10.830 --> 00:33:13.470 align:middle line:84% of pooling and some nonlinearities. 00:33:13.470 --> 00:33:16.730 align:middle line:84% And then that's the input to some convolutional layers. 00:33:16.730 --> 00:33:19.818 align:middle line:84% And then these things called LSTMs, long short-term memory-- 00:33:19.818 --> 00:33:21.610 align:middle line:84% the details of this aren't really critical. 00:33:21.610 --> 00:33:23.950 align:middle line:84% The point is it's a system that was optimized 00:33:23.950 --> 00:33:25.990 align:middle line:90% in order to recognize speech. 00:33:25.990 --> 00:33:29.630 align:middle line:84% And part of that optimization involved optimizing 00:33:29.630 --> 00:33:32.212 align:middle line:90% this front-end filter bank. 00:33:32.212 --> 00:33:33.170 align:middle line:90% And so you can do that. 00:33:33.170 --> 00:33:36.230 align:middle line:84% And then you can ask, well, what kind of filters 00:33:36.230 --> 00:33:39.710 align:middle line:84% emerge as a good engineering solution 00:33:39.710 --> 00:33:41.930 align:middle line:84% to the problem of recognizing speech? 00:33:41.930 --> 00:33:46.245 align:middle line:90% 00:33:46.245 --> 00:33:49.990 align:middle line:84% So this is one way of looking at that. 00:33:49.990 --> 00:33:54.800 align:middle line:84% So these graphs are plotting the set 00:33:54.800 --> 00:33:57.200 align:middle line:84% of filters that are in a standard model 00:33:57.200 --> 00:34:00.852 align:middle line:84% of the cochlea that's based on biological observations, 00:34:00.852 --> 00:34:03.060 align:middle line:84% and then the filters that were learned by the system. 00:34:03.060 --> 00:34:07.840 align:middle line:84% So let's just focus on the model of the biological cochlea first. 00:34:07.840 --> 00:34:10.060 align:middle line:84% So these are the filters ordered. 00:34:10.060 --> 00:34:12.800 align:middle line:84% And there's about 40 of them here. 00:34:12.800 --> 00:34:16.060 align:middle line:84% And you can see the frequency response of each filter. 00:34:16.060 --> 00:34:17.900 align:middle line:90% So this is frequency axis. 00:34:17.900 --> 00:34:22.440 align:middle line:84% And the gray scale here plots the response of that filter 00:34:22.440 --> 00:34:25.400 align:middle line:90% to the corresponding frequency. 00:34:25.400 --> 00:34:30.639 align:middle line:84% And the dashed green line is a function 00:34:30.639 --> 00:34:34.159 align:middle line:84% that's been fit to lots of observations from the masking 00:34:34.159 --> 00:34:35.860 align:middle line:84% experiments that we just talked about. 00:34:35.860 --> 00:34:39.199 align:middle line:84% And so what you can see is that the filters in the ear 00:34:39.199 --> 00:34:41.971 align:middle line:84% are arranged from low to high frequencies 00:34:41.971 --> 00:34:43.179 align:middle line:90% with this particular mapping. 00:34:43.179 --> 00:34:45.400 align:middle line:84% But you can also see the thing that I just 00:34:45.400 --> 00:34:48.679 align:middle line:84% commented on earlier, which is that the high frequency 00:34:48.679 --> 00:34:53.090 align:middle line:84% filters are wider than the low frequency filters. 00:34:53.090 --> 00:34:56.090 align:middle line:84% So the point is that there's a wider black region 00:34:56.090 --> 00:34:58.250 align:middle line:90% here than here. 00:34:58.250 --> 00:35:00.970 align:middle line:84% So that's the same thing that we saw 00:35:00.970 --> 00:35:07.570 align:middle line:84% on this graph, where we saw that the bandwidth grew 00:35:07.570 --> 00:35:09.070 align:middle line:90% with center frequency. 00:35:09.070 --> 00:35:13.107 align:middle line:90% 00:35:13.107 --> 00:35:14.690 align:middle line:84% And so the interesting result is that, 00:35:14.690 --> 00:35:17.450 align:middle line:84% if you look at the filters that emerged 00:35:17.450 --> 00:35:21.170 align:middle line:84% as a consequence of optimizing this system to recognize speech, 00:35:21.170 --> 00:35:23.610 align:middle line:84% if you arrange them in the same way-- 00:35:23.610 --> 00:35:25.610 align:middle line:84% I mean, it's a little bit messy, but you can see 00:35:25.610 --> 00:35:26.993 align:middle line:90% some traces of the same thing. 00:35:26.993 --> 00:35:28.410 align:middle line:84% So if you look at the filters that 00:35:28.410 --> 00:35:29.650 align:middle line:84% are tuned to low frequencies, they 00:35:29.650 --> 00:35:31.730 align:middle line:84% tend to be pretty narrow, whereas the ones that 00:35:31.730 --> 00:35:34.370 align:middle line:84% are tuned for high frequencies tend to be broader. 00:35:34.370 --> 00:35:37.347 align:middle line:84% And the distribution of frequencies 00:35:37.347 --> 00:35:38.430 align:middle line:90% is a little bit different. 00:35:38.430 --> 00:35:40.890 align:middle line:84% So you can compare the green curve here to the black curve 00:35:40.890 --> 00:35:41.390 align:middle line:90% here. 00:35:41.390 --> 00:35:43.410 align:middle line:90% It doesn't fall exactly on it. 00:35:43.410 --> 00:35:47.900 align:middle line:84% But it's not a million miles away either. 00:35:47.900 --> 00:35:48.400 align:middle line:90% All right. 00:35:48.400 --> 00:35:51.820 align:middle line:84% So, so the inference that we might make from this 00:35:51.820 --> 00:35:54.900 align:middle line:84% is that the filters that are in our cochlea 00:35:54.900 --> 00:35:59.100 align:middle line:84% might be a good way to extract information from sound. 00:35:59.100 --> 00:36:02.562 align:middle line:90% Any questions about that? 00:36:02.562 --> 00:36:04.020 align:middle line:84% STUDENT: What are the alternatives, 00:36:04.020 --> 00:36:06.420 align:middle line:84% like if you optimize it to do some different task, will 00:36:06.420 --> 00:36:09.420 align:middle line:84% it show a very different kind of-- 00:36:09.420 --> 00:36:13.060 align:middle line:84% JOSH MCDERMOTT: Yeah, that's a good question. 00:36:13.060 --> 00:36:16.100 align:middle line:84% This hasn't been looked at exhaustively. 00:36:16.100 --> 00:36:19.700 align:middle line:84% I can tell you that we've done a little bit of this in our lab 00:36:19.700 --> 00:36:20.920 align:middle line:90% with some tasks. 00:36:20.920 --> 00:36:24.920 align:middle line:84% And you can get-- if the task is narrow enough, 00:36:24.920 --> 00:36:29.420 align:middle line:84% you can get some unusual solutions that come out. 00:36:29.420 --> 00:36:34.540 align:middle line:84% And so I think it doesn't have to turn out this way. 00:36:34.540 --> 00:36:36.780 align:middle line:84% You could build a filter bank that 00:36:36.780 --> 00:36:40.060 align:middle line:84% wouldn't look like this at all, where you could build one that 00:36:40.060 --> 00:36:42.340 align:middle line:84% would just along the diagonal, where everything would 00:36:42.340 --> 00:36:44.900 align:middle line:90% be exactly the same bandwidth. 00:36:44.900 --> 00:36:47.870 align:middle line:84% That's just not what emerges from the optimization. 00:36:47.870 --> 00:36:50.150 align:middle line:90% 00:36:50.150 --> 00:36:50.650 align:middle line:90% Yeah. 00:36:50.650 --> 00:36:51.770 align:middle line:90% Any other questions? 00:36:51.770 --> 00:36:54.810 align:middle line:90% 00:36:54.810 --> 00:36:55.310 align:middle line:90% OK. 00:36:55.310 --> 00:36:59.630 align:middle line:90% 00:36:59.630 --> 00:37:03.230 align:middle line:84% Now, we talked very briefly last time 00:37:03.230 --> 00:37:07.270 align:middle line:84% about how the tuning that you see in the ear 00:37:07.270 --> 00:37:09.230 align:middle line:90% changes with sound intensity. 00:37:09.230 --> 00:37:12.230 align:middle line:84% So in this respect, it's nonlinear. 00:37:12.230 --> 00:37:15.270 align:middle line:84% And this is another couple graphs 00:37:15.270 --> 00:37:16.850 align:middle line:90% that show the same of thing. 00:37:16.850 --> 00:37:21.190 align:middle line:84% So cochlear filter shapes, they change with sound level. 00:37:21.190 --> 00:37:23.730 align:middle line:84% And we'll talk in some detail about why that happens. 00:37:23.730 --> 00:37:26.250 align:middle line:84% But these are just two measurements that reflect this. 00:37:26.250 --> 00:37:28.170 align:middle line:90% So this is physiological data. 00:37:28.170 --> 00:37:31.070 align:middle line:84% This is actually made from an animal 00:37:31.070 --> 00:37:34.110 align:middle line:84% cochlea, where you cut a tiny little hole in the cochlea. 00:37:34.110 --> 00:37:36.630 align:middle line:84% And then you can look in at the motion of the cochlea 00:37:36.630 --> 00:37:38.250 align:middle line:90% in response to sound. 00:37:38.250 --> 00:37:42.670 align:middle line:84% So you measure the motion of the cochlea as a function 00:37:42.670 --> 00:37:45.040 align:middle line:90% of frequency and intensity. 00:37:45.040 --> 00:37:47.000 align:middle line:84% And the results of the experiment 00:37:47.000 --> 00:37:49.400 align:middle line:84% here are expressed in terms of gain. 00:37:49.400 --> 00:37:54.160 align:middle line:84% So the gain is the ratio between the intensity of the stimulus 00:37:54.160 --> 00:37:57.760 align:middle line:84% and the mechanical motion of the basilar membrane. 00:37:57.760 --> 00:38:02.600 align:middle line:84% So it's a measure of how much the ear is amplifying sound. 00:38:02.600 --> 00:38:06.160 align:middle line:84% And what this is showing is that, when the sound intensity 00:38:06.160 --> 00:38:08.700 align:middle line:84% is low-- so you can see the sound intensity is labeled here. 00:38:08.700 --> 00:38:10.340 align:middle line:90% That's what these numbers are. 00:38:10.340 --> 00:38:12.720 align:middle line:84% So that's 5 and 10 dB, 20, 30, 40. 00:38:12.720 --> 00:38:15.880 align:middle line:84% And so at low intensities, you get this very, very narrow 00:38:15.880 --> 00:38:20.240 align:middle line:84% tuning, where the gain is very high at a particular frequency 00:38:20.240 --> 00:38:20.740 align:middle line:90% range. 00:38:20.740 --> 00:38:22.060 align:middle line:90% This is frequency. 00:38:22.060 --> 00:38:23.960 align:middle line:90% And then it falls off. 00:38:23.960 --> 00:38:27.280 align:middle line:84% Now, as the intensity of the sound increases, 00:38:27.280 --> 00:38:29.880 align:middle line:90% the gain overall drops. 00:38:29.880 --> 00:38:33.360 align:middle line:84% But you can also see that the tuning gets broader. 00:38:33.360 --> 00:38:36.880 align:middle line:84% And so down by the time that you're at 80 decibels, 00:38:36.880 --> 00:38:43.418 align:middle line:84% the tuning is much, much broader than it was at 30 decibels. 00:38:43.418 --> 00:38:44.710 align:middle line:90% What do these pictures show us? 00:38:44.710 --> 00:38:49.650 align:middle line:84% Well, they show us that the ear is providing amplification. 00:38:49.650 --> 00:38:52.130 align:middle line:84% And we believe that that's supplied by the outer hair 00:38:52.130 --> 00:38:53.290 align:middle line:90% cells. 00:38:53.290 --> 00:38:54.970 align:middle line:90% And it's level-dependent. 00:38:54.970 --> 00:38:59.630 align:middle line:84% So when your ear gets a very quiet sound, 00:38:59.630 --> 00:39:02.250 align:middle line:84% it's amplifying the heck out of it. 00:39:02.250 --> 00:39:04.430 align:middle line:84% When it gets a much louder sound, 00:39:04.430 --> 00:39:06.610 align:middle line:90% it's amplifying it much less. 00:39:06.610 --> 00:39:11.110 align:middle line:84% And it's doing that in a way that is frequency-dependent. 00:39:11.110 --> 00:39:14.970 align:middle line:84% And so the consequence of that is that, at low intensities, 00:39:14.970 --> 00:39:17.330 align:middle line:84% the tuning tends to be pretty narrow. 00:39:17.330 --> 00:39:22.210 align:middle line:84% And at higher intensities, it gets broader. 00:39:22.210 --> 00:39:24.570 align:middle line:84% So this is a physiological measurement, again, 00:39:24.570 --> 00:39:27.530 align:middle line:84% made of the actual motion of the cochlea. 00:39:27.530 --> 00:39:30.410 align:middle line:84% This is psychophysical measurements 00:39:30.410 --> 00:39:32.210 align:middle line:84% made using the masking experiments 00:39:32.210 --> 00:39:33.810 align:middle line:90% that we've been discussing. 00:39:33.810 --> 00:39:37.610 align:middle line:84% So each one of these curves is the measurement 00:39:37.610 --> 00:39:42.670 align:middle line:84% that's inferred from masking in a human with a particular signal 00:39:42.670 --> 00:39:43.170 align:middle line:90% frequency. 00:39:43.170 --> 00:39:45.212 align:middle line:84% So this is one signal frequency, this is another, 00:39:45.212 --> 00:39:46.850 align:middle line:84% this is another, this is another. 00:39:46.850 --> 00:39:50.830 align:middle line:84% And the different curves here are measured at different SPL 00:39:50.830 --> 00:39:53.030 align:middle line:90% levels of the signal. 00:39:53.030 --> 00:39:54.730 align:middle line:90% So this is dB SPL. 00:39:54.730 --> 00:39:56.250 align:middle line:90% So that's pretty quiet. 00:39:56.250 --> 00:39:58.410 align:middle line:84% And the highest one is 70 dB SPL. 00:39:58.410 --> 00:40:01.790 align:middle line:84% And you can see that the low intensities give you 00:40:01.790 --> 00:40:04.110 align:middle line:84% narrower tuning than the high intensities. 00:40:04.110 --> 00:40:09.590 align:middle line:84% So that's a psychophysical analog of that phenomenon. 00:40:09.590 --> 00:40:10.090 align:middle line:90% All right. 00:40:10.090 --> 00:40:12.070 align:middle line:84% So in general, we see sharp tuning 00:40:12.070 --> 00:40:15.090 align:middle line:84% at low sound levels, broader tuning at high sound levels. 00:40:15.090 --> 00:40:17.510 align:middle line:84% And you can see this both physiologically and 00:40:17.510 --> 00:40:20.790 align:middle line:90% psychophysically. 00:40:20.790 --> 00:40:24.750 align:middle line:84% So another way to look at this level-dependent amplification 00:40:24.750 --> 00:40:27.910 align:middle line:84% is with a graph like this that shows 00:40:27.910 --> 00:40:31.030 align:middle line:84% the response of the cochlea, in particular the basilar 00:40:31.030 --> 00:40:34.110 align:middle line:84% membrane, as a function of sound pressure level. 00:40:34.110 --> 00:40:36.110 align:middle line:84% And when you look at it in this way, 00:40:36.110 --> 00:40:37.800 align:middle line:84% the basilar membrane exhibits what's 00:40:37.800 --> 00:40:40.900 align:middle line:84% called a compressive response function. 00:40:40.900 --> 00:40:43.280 align:middle line:90% So what does this all mean? 00:40:43.280 --> 00:40:46.240 align:middle line:84% So we've got sound pressure here on the x-axis 00:40:46.240 --> 00:40:49.320 align:middle line:84% and two different measures of the basilar membrane response, 00:40:49.320 --> 00:40:51.360 align:middle line:84% displacement and velocity, which tell 00:40:51.360 --> 00:40:54.240 align:middle line:90% the same story on the y-axis. 00:40:54.240 --> 00:40:56.420 align:middle line:84% And they're both being plotted on log scales. 00:40:56.420 --> 00:40:57.460 align:middle line:90% This is in decibels. 00:40:57.460 --> 00:41:00.320 align:middle line:90% And this is a logarithmic scale. 00:41:00.320 --> 00:41:03.680 align:middle line:84% And for reference, this line here 00:41:03.680 --> 00:41:06.280 align:middle line:84% displays the motion of the stapes. 00:41:06.280 --> 00:41:08.383 align:middle line:84% Remember, the stapes is the stirrup bone. 00:41:08.383 --> 00:41:09.800 align:middle line:84% It's the last bone in the ossicles 00:41:09.800 --> 00:41:12.160 align:middle line:90% that connects to the cochlea. 00:41:12.160 --> 00:41:14.280 align:middle line:84% So you can measure the motion of the stapes. 00:41:14.280 --> 00:41:17.320 align:middle line:84% And what you see there makes perfect sense. 00:41:17.320 --> 00:41:21.560 align:middle line:84% This is a line with a slope of 1. 00:41:21.560 --> 00:41:23.220 align:middle line:90% So that's a linear response. 00:41:23.220 --> 00:41:26.020 align:middle line:84% So if you increase the sound intensity by a certain amount, 00:41:26.020 --> 00:41:28.692 align:middle line:84% the motion of the stapes increases by the same amount. 00:41:28.692 --> 00:41:30.400 align:middle line:84% Now, what happens on the basilar membrane 00:41:30.400 --> 00:41:31.880 align:middle line:84% is something that's very different. 00:41:31.880 --> 00:41:34.890 align:middle line:90% 00:41:34.890 --> 00:41:40.210 align:middle line:84% Instead, you increase the sound pressure by a certain amount. 00:41:40.210 --> 00:41:43.690 align:middle line:84% And the difference in the motion of the basilar membrane 00:41:43.690 --> 00:41:45.570 align:middle line:90% is substantially smaller. 00:41:45.570 --> 00:41:47.150 align:middle line:90% So it's a compressive response. 00:41:47.150 --> 00:41:49.450 align:middle line:84% Now, what's actually happening here 00:41:49.450 --> 00:41:52.170 align:middle line:84% is that there is amplification that is 00:41:52.170 --> 00:41:53.950 align:middle line:90% specific to very low intensity. 00:41:53.950 --> 00:41:58.770 align:middle line:84% So when you have a sound that's down here, 10 or 20 dB, 00:41:58.770 --> 00:42:00.650 align:middle line:84% the cochlea is amplifying the response 00:42:00.650 --> 00:42:02.610 align:middle line:90% to a pretty enormous extent. 00:42:02.610 --> 00:42:05.130 align:middle line:84% And one way to think about this is 00:42:05.130 --> 00:42:07.690 align:middle line:84% to compare the response at the stapes 00:42:07.690 --> 00:42:10.310 align:middle line:84% to the response on the basilar membrane. 00:42:10.310 --> 00:42:13.730 align:middle line:84% So at high levels, you can see that the response at the stapes 00:42:13.730 --> 00:42:17.210 align:middle line:84% and the response of the basilar membrane are fairly comparable. 00:42:17.210 --> 00:42:19.110 align:middle line:90% But they diverge. 00:42:19.110 --> 00:42:21.450 align:middle line:84% So the point is that, at the stapes, 00:42:21.450 --> 00:42:24.130 align:middle line:84% the response to these very low intensities 00:42:24.130 --> 00:42:25.470 align:middle line:90% is like off the graph. 00:42:25.470 --> 00:42:27.650 align:middle line:90% It's like down here. 00:42:27.650 --> 00:42:30.730 align:middle line:84% But the response of the basilar membrane is way up here. 00:42:30.730 --> 00:42:33.940 align:middle line:84% That's because you have this amplifier in your ear 00:42:33.940 --> 00:42:39.100 align:middle line:84% that's boosting the response to these low intensities. 00:42:39.100 --> 00:42:42.920 align:middle line:84% And it doesn't boost the high intensities nearly as much. 00:42:42.920 --> 00:42:45.480 align:middle line:84% And that's what gives you this compressive response function. 00:42:45.480 --> 00:42:48.240 align:middle line:90% So this is called compression. 00:42:48.240 --> 00:42:51.780 align:middle line:84% But really, it's level-dependent amplification. 00:42:51.780 --> 00:42:53.820 align:middle line:84% There's an amplifier in your ear. 00:42:53.820 --> 00:42:57.900 align:middle line:84% And it acts specifically when the sounds are low in intensity. 00:42:57.900 --> 00:43:02.140 align:middle line:84% And again, we believe this is the outer hair cells, which 00:43:02.140 --> 00:43:05.540 align:middle line:84% provide this kind of feedback system, whereby 00:43:05.540 --> 00:43:08.240 align:middle line:84% they get a little bit of input, and then they move, 00:43:08.240 --> 00:43:10.880 align:middle line:84% which, in turn, increases the motion of the basilar membrane. 00:43:10.880 --> 00:43:15.380 align:middle line:84% It's a little amplifier in your ear. 00:43:15.380 --> 00:43:15.880 align:middle line:90% All right. 00:43:15.880 --> 00:43:20.580 align:middle line:84% Now, the story is a little bit more complicated than that. 00:43:20.580 --> 00:43:24.740 align:middle line:84% And that is that the amplification really only occurs 00:43:24.740 --> 00:43:27.900 align:middle line:84% for frequencies that are near the preferred frequency 00:43:27.900 --> 00:43:29.620 align:middle line:90% of a place on the cochlea. 00:43:29.620 --> 00:43:33.110 align:middle line:84% So these are, again, measurements 00:43:33.110 --> 00:43:34.458 align:middle line:90% of the basilar membrane motion. 00:43:34.458 --> 00:43:36.750 align:middle line:84% So it's the same kind of graph we were just looking at. 00:43:36.750 --> 00:43:39.310 align:middle line:84% But now, the different lines here-- and this 00:43:39.310 --> 00:43:41.930 align:middle line:84% is all made at one place on the cochlea. 00:43:41.930 --> 00:43:43.390 align:middle line:84% And the different lines correspond 00:43:43.390 --> 00:43:45.830 align:middle line:90% to different frequencies. 00:43:45.830 --> 00:43:47.550 align:middle line:84% So this is made at the place that 00:43:47.550 --> 00:43:49.030 align:middle line:84% corresponds to 10 kilohertz, which 00:43:49.030 --> 00:43:53.550 align:middle line:90% is shown in the open symbols. 00:43:53.550 --> 00:43:56.550 align:middle line:84% And so at 10 kilohertz, you see this very shallow slope 00:43:56.550 --> 00:43:59.110 align:middle line:90% here for the cochlea's response. 00:43:59.110 --> 00:44:02.330 align:middle line:84% If you increase the frequency a little bit to 11 kilohertz, 00:44:02.330 --> 00:44:04.470 align:middle line:84% you still see a pretty shallow slope. 00:44:04.470 --> 00:44:07.590 align:middle line:84% But then as you go to 12 and to 13, 00:44:07.590 --> 00:44:12.030 align:middle line:84% the slope shifts to being pretty close to 1. 00:44:12.030 --> 00:44:15.310 align:middle line:84% So as the frequency moves away from 00:44:15.310 --> 00:44:18.550 align:middle line:84% the characteristic frequency, the response 00:44:18.550 --> 00:44:20.090 align:middle line:90% becomes much more linear. 00:44:20.090 --> 00:44:24.550 align:middle line:84% So you stop getting this nonlinear amplification. 00:44:24.550 --> 00:44:27.190 align:middle line:84% So this picture is actually telling you 00:44:27.190 --> 00:44:32.880 align:middle line:84% really the same thing seeing as this picture. 00:44:32.880 --> 00:44:37.120 align:middle line:84% There are two different ways of looking at the same phenomenon. 00:44:37.120 --> 00:44:39.360 align:middle line:84% So at low intensities, you're getting 00:44:39.360 --> 00:44:41.140 align:middle line:90% this massive amplification. 00:44:41.140 --> 00:44:42.937 align:middle line:84% But it's very frequency-selective. 00:44:42.937 --> 00:44:44.520 align:middle line:84% And so that's why, at low intensities, 00:44:44.520 --> 00:44:46.960 align:middle line:84% you get this very narrow tuning, because it's only 00:44:46.960 --> 00:44:49.293 align:middle line:84% the frequencies that the characteristic frequencies that 00:44:49.293 --> 00:44:50.120 align:middle line:90% get boosted. 00:44:50.120 --> 00:44:54.320 align:middle line:84% So at higher intensities, there's much less amplification 00:44:54.320 --> 00:44:55.600 align:middle line:90% happening. 00:44:55.600 --> 00:44:59.500 align:middle line:84% And as a consequence, the tuning gets much, much broader. 00:44:59.500 --> 00:45:04.272 align:middle line:90% 00:45:04.272 --> 00:45:06.840 align:middle line:90% Any questions about this? 00:45:06.840 --> 00:45:09.002 align:middle line:84% You're going to have to go over this a few times. 00:45:09.002 --> 00:45:11.460 align:middle line:84% It just takes a little while to wrap your head around this. 00:45:11.460 --> 00:45:13.280 align:middle line:84% And so there's going to be a problem 00:45:13.280 --> 00:45:15.447 align:middle line:84% on the problem set where you think a little bit more 00:45:15.447 --> 00:45:16.200 align:middle line:90% about this. 00:45:16.200 --> 00:45:20.900 align:middle line:84% But the big picture here is that the ear is nonlinear. 00:45:20.900 --> 00:45:23.680 align:middle line:84% You've got this amplifier in your ear. 00:45:23.680 --> 00:45:27.400 align:middle line:84% And that is helping you hear these very low intensity sounds 00:45:27.400 --> 00:45:29.630 align:middle line:84% that we have to deal with on a daily basis. 00:45:29.630 --> 00:45:30.850 align:middle line:90% Yeah? 00:45:30.850 --> 00:45:32.810 align:middle line:84% STUDENT: What is the characteristic frequency 00:45:32.810 --> 00:45:33.890 align:middle line:90% [INAUDIBLE]? 00:45:33.890 --> 00:45:35.970 align:middle line:84% JOSH MCDERMOTT: So the characteristic frequency 00:45:35.970 --> 00:45:39.810 align:middle line:84% is the frequency to which a given place on the cochlea 00:45:39.810 --> 00:45:43.410 align:middle line:84% is most sensitive, responds the best. 00:45:43.410 --> 00:45:48.279 align:middle line:90% STUDENT: [INAUDIBLE] 00:45:48.279 --> 00:45:51.550 align:middle line:84% JOSH MCDERMOTT: In this graph or in this graph? 00:45:51.550 --> 00:45:55.450 align:middle line:90% It would be 10 kilohertz. 00:45:55.450 --> 00:45:58.690 align:middle line:84% So it's like you're getting the biggest response here 00:45:58.690 --> 00:46:02.710 align:middle line:90% at 10 kilohertz. 00:46:02.710 --> 00:46:06.070 align:middle line:84% So on this graph, it would be it would be 10 kilohertz. 00:46:06.070 --> 00:46:09.410 align:middle line:84% On this graph, each one of these fibers 00:46:09.410 --> 00:46:11.850 align:middle line:84% has a different characteristic frequency. 00:46:11.850 --> 00:46:16.130 align:middle line:84% It gives you the tip of the tuning curve. 00:46:16.130 --> 00:46:19.130 align:middle line:84% And it corresponds to the place along the cochlea. 00:46:19.130 --> 00:46:19.910 align:middle line:90% Yeah, go ahead. 00:46:19.910 --> 00:46:21.170 align:middle line:90% STUDENT: [INAUDIBLE] 00:46:21.170 --> 00:46:28.110 align:middle line:90% 00:46:28.110 --> 00:46:30.910 align:middle line:84% JOSH MCDERMOTT: Ah, that I don't-- 00:46:30.910 --> 00:46:33.190 align:middle line:84% you mean like with this converged 00:46:33.190 --> 00:46:36.670 align:middle line:84% to being right on the dashed line? 00:46:36.670 --> 00:46:37.495 align:middle line:90% I'm not sure. 00:46:37.495 --> 00:46:38.870 align:middle line:84% You might have-- there's probably 00:46:38.870 --> 00:46:40.670 align:middle line:84% some baseline level of amplification 00:46:40.670 --> 00:46:42.790 align:middle line:90% that you get throughout. 00:46:42.790 --> 00:46:44.450 align:middle line:84% Yeah, I'm not positive about that. 00:46:44.450 --> 00:46:48.190 align:middle line:84% But it gets pretty close, as you can see, yeah. 00:46:48.190 --> 00:46:49.490 align:middle line:90% Yeah, any other questions? 00:46:49.490 --> 00:46:52.670 align:middle line:90% 00:46:52.670 --> 00:46:53.170 align:middle line:90% All right. 00:46:53.170 --> 00:46:57.630 align:middle line:84% Now, one of the consequences of the fact that you 00:46:57.630 --> 00:47:00.790 align:middle line:84% have these nonlinearities in your ear 00:47:00.790 --> 00:47:03.630 align:middle line:84% is that your ear creates frequencies 00:47:03.630 --> 00:47:08.150 align:middle line:84% that are not necessarily present in the input to the ear. 00:47:08.150 --> 00:47:12.460 align:middle line:84% And these are called distortion products. 00:47:12.460 --> 00:47:17.110 align:middle line:84% So the idea is that sinusoids are eigenfunctions 00:47:17.110 --> 00:47:18.550 align:middle line:90% of a linear system. 00:47:18.550 --> 00:47:21.270 align:middle line:84% So that means that, if you have a linear system, if you 00:47:21.270 --> 00:47:24.800 align:middle line:84% put in two frequencies, the output of the system 00:47:24.800 --> 00:47:27.400 align:middle line:84% will contain those same two frequencies, potentially scaled 00:47:27.400 --> 00:47:29.840 align:middle line:84% in amplitude and shifted in phase. 00:47:29.840 --> 00:47:32.840 align:middle line:84% But if the system is nonlinear, then you 00:47:32.840 --> 00:47:35.680 align:middle line:84% can get additional frequencies that come out. 00:47:35.680 --> 00:47:37.720 align:middle line:84% And that's, in fact, what happens in the ear. 00:47:37.720 --> 00:47:41.880 align:middle line:84% So here's a situation where we've got f1 and f2. 00:47:41.880 --> 00:47:45.120 align:middle line:84% So we're supplying these two frequencies to the ear. 00:47:45.120 --> 00:47:47.280 align:middle line:84% And there are these other frequencies 00:47:47.280 --> 00:47:49.000 align:middle line:84% that we might potentially observe 00:47:49.000 --> 00:47:52.700 align:middle line:84% depending on the type of nonlinearity that's in the ear. 00:47:52.700 --> 00:47:54.480 align:middle line:84% So if it's a quadratic nonlinearity, 00:47:54.480 --> 00:47:55.880 align:middle line:90% it involves a squaring. 00:47:55.880 --> 00:47:57.340 align:middle line:90% You'd get f1 minus f2. 00:47:57.340 --> 00:48:01.480 align:middle line:84% If it's a cubic nonlinearity, you'd get 2 f1 minus f2. 00:48:01.480 --> 00:48:03.520 align:middle line:84% And so you can think of any nonlinearity 00:48:03.520 --> 00:48:06.140 align:middle line:84% as being approximated by a Taylor series. 00:48:06.140 --> 00:48:08.420 align:middle line:84% So there'd be a quadratic term and a cubic term. 00:48:08.420 --> 00:48:15.620 align:middle line:84% And those tend to be pretty prominent in our ears. 00:48:15.620 --> 00:48:18.120 align:middle line:84% So we would expect that, if you've 00:48:18.120 --> 00:48:20.580 align:middle line:84% got this like powerful nonlinearity in your ear, 00:48:20.580 --> 00:48:24.930 align:middle line:84% it's going to cause these new frequencies to be present. 00:48:24.930 --> 00:48:27.770 align:middle line:84% How would we potentially detect these? 00:48:27.770 --> 00:48:29.750 align:middle line:84% So this is a classic demonstration. 00:48:29.750 --> 00:48:32.010 align:middle line:84% That's one of the main ways in which we actually 00:48:32.010 --> 00:48:33.323 align:middle line:90% measure distortion products. 00:48:33.323 --> 00:48:35.490 align:middle line:84% And it's going to make use of beating, which we just 00:48:35.490 --> 00:48:38.170 align:middle line:84% talked about at the start of the lecture. 00:48:38.170 --> 00:48:38.870 align:middle line:90% All right. 00:48:38.870 --> 00:48:41.690 align:middle line:84% And we'll listen to this demonstration. 00:48:41.690 --> 00:48:43.270 align:middle line:84% But I'm going to explain it first. 00:48:43.270 --> 00:48:47.290 align:middle line:84% So we're going to play you a sound that contains two pure 00:48:47.290 --> 00:48:51.210 align:middle line:84% tones, one at 1,000 hertz and one at 1,200 hertz. 00:48:51.210 --> 00:48:55.890 align:middle line:84% So the cubic distortion product, which we would predict 00:48:55.890 --> 00:48:57.650 align:middle line:84% would be present in your ear if there 00:48:57.650 --> 00:49:02.050 align:middle line:84% is a cubic component to the nonlinear function in your ear-- 00:49:02.050 --> 00:49:06.010 align:middle line:84% the cubic distortion product would be 2 times f1 minus f2. 00:49:06.010 --> 00:49:14.072 align:middle line:84% So 2 times f1 would be 2,000 minus f2 is 800 hertz. 00:49:14.072 --> 00:49:15.530 align:middle line:84% So the theory predicts there should 00:49:15.530 --> 00:49:18.530 align:middle line:84% be this additional frequency that's being created in your ear 00:49:18.530 --> 00:49:22.660 align:middle line:84% by its nonlinear properties at 800 hertz. 00:49:22.660 --> 00:49:26.220 align:middle line:84% So to probe for the presence of that, 00:49:26.220 --> 00:49:30.540 align:middle line:84% we're going to add a pure tone at 804 hertz. 00:49:30.540 --> 00:49:34.780 align:middle line:84% Now, the logic here is that, if your ear is creating 800 hertz 00:49:34.780 --> 00:49:38.700 align:middle line:84% and we add an 804 hertz, there's going to be beating. 00:49:38.700 --> 00:49:40.620 align:middle line:84% Now, what I didn't tell you before 00:49:40.620 --> 00:49:44.260 align:middle line:84% is that beating always happens at the difference of two 00:49:44.260 --> 00:49:46.300 align:middle line:90% frequencies. 00:49:46.300 --> 00:49:49.340 align:middle line:84% So if we've got 800 hertz and 804 hertz, 00:49:49.340 --> 00:49:52.900 align:middle line:84% there should be beating at 4 hertz, four times a second. 00:49:52.900 --> 00:49:58.535 align:middle line:84% So you'll hear this low frequency fluctuating wobble. 00:49:58.535 --> 00:50:00.160 align:middle line:84% And so we're going to actually do this. 00:50:00.160 --> 00:50:03.830 align:middle line:84% And you can see whether you hear this. 00:50:03.830 --> 00:50:05.580 align:middle line:84% And I'm going to show you this, so you can 00:50:05.580 --> 00:50:07.993 align:middle line:90% see what you're listening to. 00:50:07.993 --> 00:50:08.660 align:middle line:90% [AUDIO PLAYBACK] 00:50:08.660 --> 00:50:11.020 align:middle line:90% - Aural combination tones-- 00:50:11.020 --> 00:50:15.060 align:middle line:84% in this demonstration, two tones of 1,000 and 1,200 hertz are 00:50:15.060 --> 00:50:16.900 align:middle line:90% presented. 00:50:16.900 --> 00:50:20.310 align:middle line:84% When an 804 hertz probe tone is added, 00:50:20.310 --> 00:50:25.204 align:middle line:84% it beats with the 800 hertz aural combination tone. 00:50:25.204 --> 00:50:31.132 align:middle line:90% [AURAL COMBINATION TONES] 00:50:31.132 --> 00:50:45.458 align:middle line:90% 00:50:45.458 --> 00:50:46.630 align:middle line:90% [END PLAYBACK] 00:50:46.630 --> 00:50:48.630 align:middle line:84% JOSH MCDERMOTT: All right, so the point is that, 00:50:48.630 --> 00:50:53.710 align:middle line:84% when you just have the two primaries present at 1,000 00:50:53.710 --> 00:50:56.530 align:middle line:84% and 1,200, you shouldn't hear this subtle fluctuation. 00:50:56.530 --> 00:51:00.170 align:middle line:84% Then we add in the third tone at 804 hertz 00:51:00.170 --> 00:51:02.890 align:middle line:84% and you should hear this subtle pulsation. 00:51:02.890 --> 00:51:05.230 align:middle line:84% So that's how you're supposed to tell whether you've got 00:51:05.230 --> 00:51:06.530 align:middle line:90% the distortion product or not. 00:51:06.530 --> 00:51:09.390 align:middle line:90% So we can just try it again. 00:51:09.390 --> 00:51:10.376 align:middle line:90% [AUDIO PLAYBACK] 00:51:10.376 --> 00:51:15.306 align:middle line:90% [AURAL COMBINATION TONES] 00:51:15.306 --> 00:51:32.068 align:middle line:90% 00:51:32.068 --> 00:51:32.937 align:middle line:90% [END PLAYBACK] 00:51:32.937 --> 00:51:34.020 align:middle line:90% JOSH MCDERMOTT: All right. 00:51:34.020 --> 00:51:36.720 align:middle line:84% So it's pretty subtle, but it's there. 00:51:36.720 --> 00:51:41.302 align:middle line:84% And in a second, I'll play you something that's not as subtle. 00:51:41.302 --> 00:51:42.760 align:middle line:84% So this is actually one of the ways 00:51:42.760 --> 00:51:44.660 align:middle line:84% that people actually measure these things. 00:51:44.660 --> 00:51:46.960 align:middle line:84% And there's something even crazier that you do, 00:51:46.960 --> 00:51:50.400 align:middle line:84% if you actually want to infer the amplitude of the distortion 00:51:50.400 --> 00:51:51.300 align:middle line:90% product. 00:51:51.300 --> 00:51:55.220 align:middle line:84% And that is-- so you do what we just did, all right? 00:51:55.220 --> 00:51:57.840 align:middle line:84% So you've got your two primaries, and then 00:51:57.840 --> 00:51:59.680 align:middle line:90% the beating tone. 00:51:59.680 --> 00:52:04.240 align:middle line:84% And then you add in a tone at 800 hertz. 00:52:04.240 --> 00:52:08.600 align:middle line:84% And the participant has to adjust the amplitude and phase 00:52:08.600 --> 00:52:11.260 align:middle line:84% of the 800 hertz tone that you're piping into their ear. 00:52:11.260 --> 00:52:15.450 align:middle line:84% And the idea is that, if they can get it exactly out of phase 00:52:15.450 --> 00:52:18.130 align:middle line:84% with the frequency that your ear is creating, 00:52:18.130 --> 00:52:21.610 align:middle line:84% it cancels out, destructive interference. 00:52:21.610 --> 00:52:22.990 align:middle line:90% And then the beating goes away. 00:52:22.990 --> 00:52:25.730 align:middle line:84% And so you do these incredibly painful experiments, 00:52:25.730 --> 00:52:27.790 align:middle line:84% where you're twiddling these knobs, 00:52:27.790 --> 00:52:30.210 align:middle line:84% adjusting the amplitude and phase of this 800 hertz thing 00:52:30.210 --> 00:52:31.270 align:middle line:90% to make the beating go away. 00:52:31.270 --> 00:52:32.728 align:middle line:84% And so we did this once in our lab. 00:52:32.728 --> 00:52:36.410 align:middle line:84% And it's a paper that has only 1 and 1/2 participants. 00:52:36.410 --> 00:52:39.470 align:middle line:84% One of the participants was the grad student who led the study. 00:52:39.470 --> 00:52:41.070 align:middle line:84% And he did it in both of his ears. 00:52:41.070 --> 00:52:43.230 align:middle line:84% And then the other participant was another student. 00:52:43.230 --> 00:52:45.063 align:middle line:84% And he could only bear to do one of his ears 00:52:45.063 --> 00:52:47.630 align:middle line:84% because it's really, really draining to do. 00:52:47.630 --> 00:52:48.770 align:middle line:90% But you can do it. 00:52:48.770 --> 00:52:50.490 align:middle line:84% And you can measure these things. 00:52:50.490 --> 00:52:53.610 align:middle line:84% But we have an even cooler demonstration 00:52:53.610 --> 00:52:57.490 align:middle line:84% of this, which is that we're going to-- 00:52:57.490 --> 00:53:01.370 align:middle line:84% let me turn down the volume here. 00:53:01.370 --> 00:53:01.870 align:middle line:90% All right. 00:53:01.870 --> 00:53:03.690 align:middle line:84% So we expect that there is a distortion 00:53:03.690 --> 00:53:08.610 align:middle line:84% product that's 2 f1 minus f2, the cubic distortion product. 00:53:08.610 --> 00:53:11.330 align:middle line:84% And so the idea is that, because there's a minus sign in front 00:53:11.330 --> 00:53:15.390 align:middle line:84% of the f2, if we actually induce a frequency sweep-- 00:53:15.390 --> 00:53:17.750 align:middle line:90% so we cause f2 to sweep up-- 00:53:17.750 --> 00:53:20.842 align:middle line:84% the distortion product should sweep down in frequency. 00:53:20.842 --> 00:53:22.550 align:middle line:84% And so that actually makes it really easy 00:53:22.550 --> 00:53:24.790 align:middle line:84% to hear because you can tell that there's 00:53:24.790 --> 00:53:28.030 align:middle line:84% this thing that you're hearing that's going down 00:53:28.030 --> 00:53:31.630 align:middle line:84% when the other stuff is either staying stationary or going up. 00:53:31.630 --> 00:53:33.750 align:middle line:90% So this is what you'll hear. 00:53:33.750 --> 00:53:37.070 align:middle line:90% [F1 TONE] 00:53:37.070 --> 00:53:38.570 align:middle line:90% So that's f1. 00:53:38.570 --> 00:53:39.302 align:middle line:90% [F2 TONE] 00:53:39.302 --> 00:53:42.110 align:middle line:90% That's f2. 00:53:42.110 --> 00:53:44.270 align:middle line:90% And now, the both. 00:53:44.270 --> 00:53:49.710 align:middle line:90% [F1 AND F2 TONES] 00:53:49.710 --> 00:53:51.150 align:middle line:90% You hear it going down? 00:53:51.150 --> 00:53:53.190 align:middle line:90% OK. 00:53:53.190 --> 00:53:54.790 align:middle line:90% OK, let's do it again. 00:53:54.790 --> 00:54:01.350 align:middle line:90% [F1 AND F2 TONES] 00:54:01.350 --> 00:54:04.430 align:middle line:84% So you can see how the stimulus is going up. 00:54:04.430 --> 00:54:06.390 align:middle line:84% That thing that was going down is in your ear. 00:54:06.390 --> 00:54:07.570 align:middle line:90% It's not in the signal. 00:54:07.570 --> 00:54:11.260 align:middle line:84% So that's a distortion product that your ear is creating. 00:54:11.260 --> 00:54:15.400 align:middle line:90% 00:54:15.400 --> 00:54:18.920 align:middle line:84% Any questions about distortion products? 00:54:18.920 --> 00:54:19.520 align:middle line:90% Yeah? 00:54:19.520 --> 00:54:20.978 align:middle line:84% STUDENT: I don't understand what it 00:54:20.978 --> 00:54:23.660 align:middle line:84% means to have something created in the ear. 00:54:23.660 --> 00:54:25.060 align:middle line:90% Is it like a perceived thing? 00:54:25.060 --> 00:54:28.420 align:middle line:84% Are we creating it not in our minds? 00:54:28.420 --> 00:54:32.437 align:middle line:84% Like, how does it have physical tangibility in the ear? 00:54:32.437 --> 00:54:34.520 align:middle line:84% JOSH MCDERMOTT: Well, yeah, I use the word created 00:54:34.520 --> 00:54:39.300 align:middle line:84% because the nonlinearity in your ear, 00:54:39.300 --> 00:54:43.680 align:middle line:84% it actually is reflected in the motion of the basilar membrane. 00:54:43.680 --> 00:54:46.040 align:middle line:84% So your basilar membrane will actually 00:54:46.040 --> 00:54:49.607 align:middle line:84% be vibrating at the frequency of the distortion product. 00:54:49.607 --> 00:54:50.440 align:middle line:90% STUDENT: [INAUDIBLE] 00:54:50.440 --> 00:54:51.960 align:middle line:84% JOSH MCDERMOTT: That's correct, yeah. 00:54:51.960 --> 00:54:52.460 align:middle line:90% Yeah. 00:54:52.460 --> 00:54:55.080 align:middle line:90% 00:54:55.080 --> 00:54:58.640 align:middle line:84% Yeah, any other questions about that? 00:54:58.640 --> 00:55:01.300 align:middle line:90% OK. 00:55:01.300 --> 00:55:05.753 align:middle line:84% So take-home messages are the ear is frequency selective. 00:55:05.753 --> 00:55:07.420 align:middle line:84% You can't measure it directly in humans. 00:55:07.420 --> 00:55:09.070 align:middle line:84% You have to infer it from masking. 00:55:09.070 --> 00:55:11.403 align:middle line:84% So we talked a little bit about how you do those masking 00:55:11.403 --> 00:55:12.150 align:middle line:90% experiments. 00:55:12.150 --> 00:55:14.290 align:middle line:84% The results of the masking experiments 00:55:14.290 --> 00:55:18.350 align:middle line:84% show that frequency selectivity varies with frequency. 00:55:18.350 --> 00:55:20.430 align:middle line:84% So at high frequencies, it tends to be broader 00:55:20.430 --> 00:55:21.430 align:middle line:90% than at low frequencies. 00:55:21.430 --> 00:55:23.850 align:middle line:84% We talked about some evidence that maybe that's a good way 00:55:23.850 --> 00:55:27.930 align:middle line:84% to design an ear in order to help you perform auditory tasks. 00:55:27.930 --> 00:55:30.570 align:middle line:84% We also talked about how frequency selectivity changes 00:55:30.570 --> 00:55:31.630 align:middle line:90% with level. 00:55:31.630 --> 00:55:34.090 align:middle line:84% And that's because there is amplification that's 00:55:34.090 --> 00:55:38.250 align:middle line:84% happening in your ears that is both level-dependent and 00:55:38.250 --> 00:55:39.890 align:middle line:90% frequency-specific. 00:55:39.890 --> 00:55:43.130 align:middle line:84% So it happens at low sound levels 00:55:43.130 --> 00:55:46.690 align:middle line:84% at frequencies that are close to the characteristic frequency. 00:55:46.690 --> 00:55:49.690 align:middle line:84% And the consequence of that is that your ear 00:55:49.690 --> 00:55:51.190 align:middle line:90% is a nonlinear system. 00:55:51.190 --> 00:55:53.610 align:middle line:84% And that is evident in these distortion products 00:55:53.610 --> 00:55:56.122 align:middle line:84% that you can hear if you do clever things. 00:55:56.122 --> 00:55:58.330 align:middle line:84% And so I should also emphasize about these distortion 00:55:58.330 --> 00:55:58.870 align:middle line:90% products-- 00:55:58.870 --> 00:56:00.530 align:middle line:84% I mean, these are there all the time. 00:56:00.530 --> 00:56:02.470 align:middle line:84% It's just that most of the time you don't notice them. 00:56:02.470 --> 00:56:04.553 align:middle line:84% It's just part of the way the world sounds to you. 00:56:04.553 --> 00:56:07.940 align:middle line:84% And you have to do these funny experiments in order 00:56:07.940 --> 00:56:12.700 align:middle line:84% to see very, very clear evidence of them. 00:56:12.700 --> 00:56:16.480 align:middle line:84% OK, let's start talking about hearing loss. 00:56:16.480 --> 00:56:20.070 align:middle line:90% 00:56:20.070 --> 00:56:23.940 align:middle line:84% So this is a graph that shows normal auditory sensitivity. 00:56:23.940 --> 00:56:28.300 align:middle line:84% So this is plotting detection thresholds expressed 00:56:28.300 --> 00:56:31.740 align:middle line:84% as the sound intensity that you can just barely detect 00:56:31.740 --> 00:56:33.600 align:middle line:90% as a function of frequency. 00:56:33.600 --> 00:56:36.420 align:middle line:90% So it has this U shape. 00:56:36.420 --> 00:56:38.620 align:middle line:84% So you're most sensitive typically 00:56:38.620 --> 00:56:40.620 align:middle line:90% around 2 or 3 kilohertz. 00:56:40.620 --> 00:56:42.940 align:middle line:84% And then when frequencies get lower, 00:56:42.940 --> 00:56:44.040 align:middle line:90% you become less sensitive. 00:56:44.040 --> 00:56:46.660 align:middle line:84% And when they become higher, you're less sensitive. 00:56:46.660 --> 00:56:52.340 align:middle line:84% So this is just what somebody with normal hearing looks like. 00:56:52.340 --> 00:56:56.235 align:middle line:84% This is what gets measured when you go to the doctor. 00:56:56.235 --> 00:56:57.360 align:middle line:90% You want to pass those out? 00:56:57.360 --> 00:56:57.960 align:middle line:90% Yeah, sure. 00:56:57.960 --> 00:57:00.308 align:middle line:90% STUDENT: [INAUDIBLE] 00:57:00.308 --> 00:57:02.600 align:middle line:84% JOSH MCDERMOTT: Ah, we have some hard copies of slides. 00:57:02.600 --> 00:57:04.475 align:middle line:84% There's a few people that wanted those, yeah. 00:57:04.475 --> 00:57:06.050 align:middle line:84% OK, so this is what gets measured. 00:57:06.050 --> 00:57:08.675 align:middle line:84% If you go to the audiologist and you have your hearing checked. 00:57:08.675 --> 00:57:10.310 align:middle line:90% We do this in our lab. 00:57:10.310 --> 00:57:14.070 align:middle line:84% This is me getting my audiogram measured this morning thanks 00:57:14.070 --> 00:57:17.430 align:middle line:84% to a grad student who helped me do this. 00:57:17.430 --> 00:57:20.590 align:middle line:84% And so this is what it looks like if you 00:57:20.590 --> 00:57:21.530 align:middle line:90% go to the audiologist. 00:57:21.530 --> 00:57:23.230 align:middle line:90% These are frequencies here. 00:57:23.230 --> 00:57:24.690 align:middle line:90% And this is the threshold here. 00:57:24.690 --> 00:57:27.030 align:middle line:84% And I mentioned in the previous lecture 00:57:27.030 --> 00:57:29.190 align:middle line:84% that we often talk about hearing level. 00:57:29.190 --> 00:57:31.470 align:middle line:84% So there's often SPL measurements 00:57:31.470 --> 00:57:33.470 align:middle line:84% that are absolute level measurements 00:57:33.470 --> 00:57:35.470 align:middle line:90% relative to the standard. 00:57:35.470 --> 00:57:38.510 align:middle line:84% And then hearing level is expressed relative to what's 00:57:38.510 --> 00:57:39.850 align:middle line:90% considered normal hearing. 00:57:39.850 --> 00:57:44.190 align:middle line:84% So 0 would mean that you have perfect hearing. 00:57:44.190 --> 00:57:45.990 align:middle line:90% And my hearing is not perfect. 00:57:45.990 --> 00:57:47.090 align:middle line:90% But it's not terrible. 00:57:47.090 --> 00:57:48.150 align:middle line:90% It could be worse. 00:57:48.150 --> 00:57:51.190 align:middle line:84% And you can see here that, at higher frequencies, 00:57:51.190 --> 00:57:53.710 align:middle line:84% I've got some significant threshold elevation, especially 00:57:53.710 --> 00:57:56.350 align:middle line:90% in the left ear. 00:57:56.350 --> 00:58:01.070 align:middle line:84% So at 4,000 hertz, my threshold is like 35 dB higher than 00:58:01.070 --> 00:58:05.000 align:middle line:84% presumably it was when I was born. 00:58:05.000 --> 00:58:07.520 align:middle line:84% So if you go to the doctor, this is the kind of thing 00:58:07.520 --> 00:58:08.280 align:middle line:90% that they'll do. 00:58:08.280 --> 00:58:09.640 align:middle line:84% If you're a hearing scientist, you often 00:58:09.640 --> 00:58:11.682 align:middle line:84% do this before you run somebody in an experiment, 00:58:11.682 --> 00:58:14.560 align:middle line:84% just to make sure their hearing is normal. 00:58:14.560 --> 00:58:19.120 align:middle line:84% And this is what they look like as a function of age on average. 00:58:19.120 --> 00:58:24.000 align:middle line:84% So this is that same thing, but now averaged within age groups. 00:58:24.000 --> 00:58:28.320 align:middle line:84% So we've got the threshold expressed in dB HL. 00:58:28.320 --> 00:58:30.220 align:middle line:84% So again, 0 is considered normal. 00:58:30.220 --> 00:58:32.440 align:middle line:84% And you can see that, in your 20s, 00:58:32.440 --> 00:58:35.440 align:middle line:84% which is the black curve and 30s, the red curve, 00:58:35.440 --> 00:58:37.040 align:middle line:90% life is pretty good. 00:58:37.040 --> 00:58:41.120 align:middle line:84% On average, there's very, very little threshold elevation. 00:58:41.120 --> 00:58:44.240 align:middle line:90% But as you move beyond that-- 00:58:44.240 --> 00:58:47.480 align:middle line:84% 40, 50, 60-- there starts to, on average, 00:58:47.480 --> 00:58:50.640 align:middle line:84% be some fairly significant threshold elevation. 00:58:50.640 --> 00:58:53.360 align:middle line:84% And it's just systematic with age. 00:58:53.360 --> 00:58:55.100 align:middle line:84% And it's really quite significant. 00:58:55.100 --> 00:58:56.880 align:middle line:84% So you can see that people in their 80s, 00:58:56.880 --> 00:58:59.730 align:middle line:84% like at these high frequencies, 4 and 8 kilohertz, 00:58:59.730 --> 00:59:02.870 align:middle line:84% we're talking decibels of threshold elevation, 00:59:02.870 --> 00:59:04.670 align:middle line:90% so pretty substantial. 00:59:04.670 --> 00:59:08.010 align:middle line:90% 00:59:08.010 --> 00:59:11.210 align:middle line:84% So this is a massive public health problem 00:59:11.210 --> 00:59:14.610 align:middle line:84% because it basically happens to almost everybody. 00:59:14.610 --> 00:59:16.770 align:middle line:84% And there's lots and lots of interest 00:59:16.770 --> 00:59:21.650 align:middle line:84% in trying to understand both why this occurs, how to treat it, 00:59:21.650 --> 00:59:23.670 align:middle line:84% whether to prevent it, things like that. 00:59:23.670 --> 00:59:24.730 align:middle line:90% Yeah? 00:59:24.730 --> 00:59:27.390 align:middle line:84% STUDENT: So does this graph-- as you go higher, 00:59:27.390 --> 00:59:28.850 align:middle line:90% does it gradually go up? 00:59:28.850 --> 00:59:30.850 align:middle line:84% Or does it asymptote at some point 00:59:30.850 --> 00:59:32.910 align:middle line:84% once you get like, higher [INAUDIBLE}. 00:59:32.910 --> 00:59:36.810 align:middle line:90% 00:59:36.810 --> 00:59:43.410 align:middle line:84% If you extended this frequency all the way to like 00:59:43.410 --> 00:59:46.050 align:middle line:90% 20 kilohertz does is it-- 00:59:46.050 --> 00:59:47.778 align:middle line:90% is it an asymptote? 00:59:47.778 --> 00:59:48.570 align:middle line:90% JOSH MCDERMOTT: No. 00:59:48.570 --> 00:59:49.697 align:middle line:90% So it would-- 00:59:49.697 --> 00:59:50.530 align:middle line:90% STUDENT: [INAUDIBLE] 00:59:50.530 --> 00:59:51.430 align:middle line:84% JOSH MCDERMOTT: Yeah, it'll look-- 00:59:51.430 --> 00:59:51.690 align:middle line:90% STUDENT: [INAUDIBLE] 00:59:51.690 --> 00:59:53.550 align:middle line:84% JOSH MCDERMOTT: Oh, well, you just-- 00:59:53.550 --> 00:59:56.130 align:middle line:90% that would be hard to measure. 00:59:56.130 --> 00:59:58.430 align:middle line:84% But yeah, no, you wouldn't really 00:59:58.430 --> 01:00:01.330 align:middle line:84% bother measuring anything that's more than 80 decibels 01:00:01.330 --> 01:00:04.450 align:middle line:84% because it's just not really practically very relevant. 01:00:04.450 --> 01:00:06.970 align:middle line:84% But yeah, I mean, it'll look worse. 01:00:06.970 --> 01:00:10.470 align:middle line:84% So I mean, typically if you take somebody in their 30s, 01:00:10.470 --> 01:00:16.230 align:middle line:84% they will have threshold elevation in the high teens 01:00:16.230 --> 01:00:17.370 align:middle line:90% of kilohertz. 01:00:17.370 --> 01:00:21.390 align:middle line:90% 01:00:21.390 --> 01:00:23.830 align:middle line:84% Yeah, the very high frequencies are the most vulnerable. 01:00:23.830 --> 01:00:26.330 align:middle line:84% And that's where you see the biggest threshold shifts. 01:00:26.330 --> 01:00:27.510 align:middle line:90% Did you have a question? 01:00:27.510 --> 01:00:28.885 align:middle line:84% STUDENT: Yeah, I was going to ask 01:00:28.885 --> 01:00:31.830 align:middle line:84% if there's any hearing loss that's characterized with losing 01:00:31.830 --> 01:00:35.390 align:middle line:84% lower frequency sounds, and then still being able to hear 01:00:35.390 --> 01:00:36.510 align:middle line:90% the higher frequencies. 01:00:36.510 --> 01:00:37.790 align:middle line:90% JOSH MCDERMOTT: Yes. 01:00:37.790 --> 01:00:41.030 align:middle line:84% It's more unusual and typically associated 01:00:41.030 --> 01:00:44.370 align:middle line:90% with particular genetics. 01:00:44.370 --> 01:00:46.610 align:middle line:84% But yeah, you do you do find people like that. 01:00:46.610 --> 01:00:47.110 align:middle line:90% Yeah? 01:00:47.110 --> 01:00:49.990 align:middle line:84% STUDENT: What's the deal with the 40 and 50-year-olds? 01:00:49.990 --> 01:00:52.590 align:middle line:84% JOSH MCDERMOTT: I think that's a midlife crisis. 01:00:52.590 --> 01:00:53.090 align:middle line:90% No. 01:00:53.090 --> 01:00:54.870 align:middle line:84% I mean, I think the sample size here 01:00:54.870 --> 01:00:57.380 align:middle line:90% is just-- it's not infinite. 01:00:57.380 --> 01:00:59.622 align:middle line:84% And so there are error bars on these. 01:00:59.622 --> 01:01:01.080 align:middle line:84% But yeah, we would certainly expect 01:01:01.080 --> 01:01:02.622 align:middle line:84% that this would be monotonic with age 01:01:02.622 --> 01:01:04.600 align:middle line:90% if you had enough data, yeah. 01:01:04.600 --> 01:01:05.240 align:middle line:90% Yeah? 01:01:05.240 --> 01:01:07.680 align:middle line:90% STUDENT: [INAUDIBLE] 01:01:07.680 --> 01:01:10.152 align:middle line:90% 01:01:10.152 --> 01:01:12.360 align:middle line:84% JOSH MCDERMOTT: So we'll talk about conductive losses 01:01:12.360 --> 01:01:12.940 align:middle line:90% in a second. 01:01:12.940 --> 01:01:14.340 align:middle line:90% Those are not so common. 01:01:14.340 --> 01:01:17.040 align:middle line:84% So this is really just reflecting 01:01:17.040 --> 01:01:20.160 align:middle line:84% typical age-related hearing loss that just 01:01:20.160 --> 01:01:21.680 align:middle line:90% happens to almost everybody. 01:01:21.680 --> 01:01:24.800 align:middle line:84% There are specific diseases that would have different phenotypes 01:01:24.800 --> 01:01:25.900 align:middle line:90% that are less common. 01:01:25.900 --> 01:01:26.800 align:middle line:90% Yeah? 01:01:26.800 --> 01:01:31.600 align:middle line:84% STUDENT: Is this true for people living in all places 01:01:31.600 --> 01:01:33.035 align:middle line:90% or if you're rural-- 01:01:33.035 --> 01:01:34.160 align:middle line:90% JOSH MCDERMOTT: Stay tuned. 01:01:34.160 --> 01:01:35.340 align:middle line:90% Yeah, great. 01:01:35.340 --> 01:01:38.160 align:middle line:90% That's a great question, yeah. 01:01:38.160 --> 01:01:38.660 align:middle line:90% OK. 01:01:38.660 --> 01:01:42.120 align:middle line:84% So just to give you a sense of the significance of this, 01:01:42.120 --> 01:01:44.320 align:middle line:84% this is the amount of attenuation 01:01:44.320 --> 01:01:47.360 align:middle line:84% that you would get from having earplugs in your ears 01:01:47.360 --> 01:01:49.900 align:middle line:84% or just putting your hands over your ears. 01:01:49.900 --> 01:01:53.300 align:middle line:84% So this is sound frequency and the amount of attenuation. 01:01:53.300 --> 01:01:57.210 align:middle line:84% So earplugs maybe give you 20 or 30 dB attenuation more 01:01:57.210 --> 01:01:59.690 align:middle line:84% at high frequencies than at low frequencies. 01:01:59.690 --> 01:02:02.910 align:middle line:84% Putting your hands over your ears gives you that. 01:02:02.910 --> 01:02:06.610 align:middle line:84% So this is on par with or maybe not even as extensive 01:02:06.610 --> 01:02:11.050 align:middle line:84% as what you would have if you just get older on average. 01:02:11.050 --> 01:02:12.730 align:middle line:84% So there's two kinds of hearing loss 01:02:12.730 --> 01:02:14.550 align:middle line:84% that we will typically distinguish between. 01:02:14.550 --> 01:02:17.635 align:middle line:84% I hope you don't do that the whole lecture. 01:02:17.635 --> 01:02:19.010 align:middle line:84% So there's conduction loss, which 01:02:19.010 --> 01:02:20.610 align:middle line:90% we were just talking about. 01:02:20.610 --> 01:02:23.010 align:middle line:84% So that's an impairment in the mechanical transmission 01:02:23.010 --> 01:02:25.470 align:middle line:84% of sound energy to the inner ear. 01:02:25.470 --> 01:02:29.130 align:middle line:84% So it's typically a problem with the outer or middle ear. 01:02:29.130 --> 01:02:32.530 align:middle line:84% So for instance, if the bones in your ear 01:02:32.530 --> 01:02:34.650 align:middle line:84% don't move as well as they should, 01:02:34.650 --> 01:02:37.650 align:middle line:84% you won't get as much energy into the cochlea as you should. 01:02:37.650 --> 01:02:41.130 align:middle line:90% And your hearing will be worse. 01:02:41.130 --> 01:02:43.690 align:middle line:84% So that's contrasted with sensory/neural loss, which 01:02:43.690 --> 01:02:46.730 align:middle line:84% is an impairment in the transduction of sound energy 01:02:46.730 --> 01:02:49.580 align:middle line:84% to electrical signals or in their transmission to the brain. 01:02:49.580 --> 01:02:51.830 align:middle line:84% And that's usually due to problems with the inner ear. 01:02:51.830 --> 01:02:54.780 align:middle line:90% 01:02:54.780 --> 01:02:56.860 align:middle line:84% So conduction loss can involve the puncturing 01:02:56.860 --> 01:02:59.480 align:middle line:84% of the eardrum-- for instance, via an ear infection. 01:02:59.480 --> 01:03:02.720 align:middle line:84% There's also otosclerosis, which was just asked about. 01:03:02.720 --> 01:03:04.420 align:middle line:84% So that's a disease that involves 01:03:04.420 --> 01:03:06.860 align:middle line:84% the gradual immobilization of the stapes bone 01:03:06.860 --> 01:03:07.760 align:middle line:90% and the ossicles. 01:03:07.760 --> 01:03:09.900 align:middle line:90% So it just doesn't move as well. 01:03:09.900 --> 01:03:11.500 align:middle line:90% It's inherited. 01:03:11.500 --> 01:03:13.300 align:middle line:84% And the good news is that this can usually 01:03:13.300 --> 01:03:14.720 align:middle line:90% be treated pretty effectively. 01:03:14.720 --> 01:03:16.500 align:middle line:84% So they can perform surgeries that 01:03:16.500 --> 01:03:18.960 align:middle line:90% will replace the stapes bone. 01:03:18.960 --> 01:03:22.940 align:middle line:84% And that's a pretty common surgery for an ear surgeon 01:03:22.940 --> 01:03:23.800 align:middle line:90% to perform. 01:03:23.800 --> 01:03:26.560 align:middle line:90% 01:03:26.560 --> 01:03:27.060 align:middle line:90% Yeah? 01:03:27.060 --> 01:03:28.580 align:middle line:90% STUDENT: [INAUDIBLE] 01:03:28.580 --> 01:03:31.340 align:middle line:90% 01:03:31.340 --> 01:03:32.540 align:middle line:90% JOSH MCDERMOTT: I forget. 01:03:32.540 --> 01:03:34.100 align:middle line:90% I'm not sure. 01:03:34.100 --> 01:03:37.220 align:middle line:90% Yeah, I can find out. 01:03:37.220 --> 01:03:43.140 align:middle line:84% So sensory/neural loss is the more vexing problem 01:03:43.140 --> 01:03:45.420 align:middle line:90% that most of us deal with. 01:03:45.420 --> 01:03:48.020 align:middle line:84% So one aspect of this is what's normally 01:03:48.020 --> 01:03:49.872 align:middle line:90% referred to as presbycusis. 01:03:49.872 --> 01:03:51.830 align:middle line:84% So that's the thing that was just on that graph 01:03:51.830 --> 01:03:53.910 align:middle line:84% that we were showing, which refers to the fact 01:03:53.910 --> 01:03:57.390 align:middle line:84% that, as people age, they become less sensitive and particularly 01:03:57.390 --> 01:03:59.830 align:middle line:84% less sensitive to high frequencies. 01:03:59.830 --> 01:04:01.330 align:middle line:84% So we didn't see this on the graph. 01:04:01.330 --> 01:04:03.430 align:middle line:84% But as I was alluding to, most people over 30 01:04:03.430 --> 01:04:05.910 align:middle line:84% typically can't hear above 15 kilohertz. 01:04:05.910 --> 01:04:08.830 align:middle line:84% Most people over 50 can't hear above 12 kilohertz. 01:04:08.830 --> 01:04:11.558 align:middle line:84% So did any of you-- like when maybe you were in high school, 01:04:11.558 --> 01:04:13.350 align:middle line:84% did you have those high frequency ringtones 01:04:13.350 --> 01:04:15.150 align:middle line:90% that your parents couldn't hear? 01:04:15.150 --> 01:04:16.392 align:middle line:90% Or you've heard of them? 01:04:16.392 --> 01:04:17.350 align:middle line:90% Your friends have them? 01:04:17.350 --> 01:04:18.450 align:middle line:90% Anyhow, this is a thing, right? 01:04:18.450 --> 01:04:20.750 align:middle line:84% You can program your phone to have a very high frequency 01:04:20.750 --> 01:04:23.250 align:middle line:84% ringtone, if you don't want your parents to know that you're 01:04:23.250 --> 01:04:24.130 align:middle line:90% getting phone calls. 01:04:24.130 --> 01:04:25.910 align:middle line:84% And it's because older people can't hear 01:04:25.910 --> 01:04:27.210 align:middle line:90% these really high frequencies. 01:04:27.210 --> 01:04:29.910 align:middle line:90% 01:04:29.910 --> 01:04:33.990 align:middle line:84% OK, so it's unclear what causes presbycusis. 01:04:33.990 --> 01:04:36.550 align:middle line:84% There appears to be a genetic component. 01:04:36.550 --> 01:04:38.150 align:middle line:90% And part of what makes this-- 01:04:38.150 --> 01:04:40.270 align:middle line:84% part of what makes this hard to understand 01:04:40.270 --> 01:04:42.150 align:middle line:84% is that while people are alive, it's 01:04:42.150 --> 01:04:44.810 align:middle line:84% really hard to actually see inside their ears. 01:04:44.810 --> 01:04:48.020 align:middle line:84% So you can measure the fact that somebody's hearing is worse, 01:04:48.020 --> 01:04:51.020 align:middle line:84% but you can't really look inside their ear, their auditory nerve, 01:04:51.020 --> 01:04:52.720 align:middle line:84% and actually directly observe what 01:04:52.720 --> 01:04:54.320 align:middle line:84% would be different about somebody 01:04:54.320 --> 01:04:59.240 align:middle line:84% who has poor hearing compared to somebody who has great hearing. 01:04:59.240 --> 01:05:01.760 align:middle line:84% So one of the main ways that this has been studied 01:05:01.760 --> 01:05:03.840 align:middle line:90% is what is called otopathology. 01:05:03.840 --> 01:05:06.900 align:middle line:84% This is postmortem studies of the cochlea. 01:05:06.900 --> 01:05:10.500 align:middle line:84% So when people die, if they donate their bodies to science, 01:05:10.500 --> 01:05:12.300 align:middle line:90% the cochlea can be removed. 01:05:12.300 --> 01:05:14.400 align:middle line:84% You can chop it up, put it under a microphone, 01:05:14.400 --> 01:05:16.920 align:middle line:84% and then look at it in all kinds of different ways. 01:05:16.920 --> 01:05:20.920 align:middle line:84% And so one of the classic things that has been done 01:05:20.920 --> 01:05:24.240 align:middle line:84% is to actually count the number of hair cells 01:05:24.240 --> 01:05:26.860 align:middle line:84% that are present in a person when they die. 01:05:26.860 --> 01:05:30.320 align:middle line:84% So this is an example of a slice of the cochlea. 01:05:30.320 --> 01:05:32.880 align:middle line:84% This is the organ of Corti, so basilar membrane here. 01:05:32.880 --> 01:05:35.520 align:middle line:84% And we've got outer hair cells here and an inner hair 01:05:35.520 --> 01:05:36.640 align:middle line:90% cell here. 01:05:36.640 --> 01:05:39.120 align:middle line:84% So you look at these things under a microphone 01:05:39.120 --> 01:05:40.060 align:middle line:90% and you count them up. 01:05:40.060 --> 01:05:42.480 align:middle line:84% And so these would be different slices 01:05:42.480 --> 01:05:44.760 align:middle line:90% along the cochlear spiral. 01:05:44.760 --> 01:05:49.430 align:middle line:84% And you just count the presence or absence of outer hair cells. 01:05:49.430 --> 01:05:50.370 align:middle line:90% So you see three here. 01:05:50.370 --> 01:05:52.210 align:middle line:84% And inner hair cells, you see one there. 01:05:52.210 --> 01:05:57.450 align:middle line:84% So present, absent as a function of distance from the base. 01:05:57.450 --> 01:06:00.330 align:middle line:84% In fact, our neighbors across the river at the Mount Sinai Ear 01:06:00.330 --> 01:06:03.678 align:middle line:84% Institute are a worldwide leader in this kind of thing. 01:06:03.678 --> 01:06:05.470 align:middle line:84% So they have lots and lots of data on this. 01:06:05.470 --> 01:06:07.053 align:middle line:84% And these slides are actually borrowed 01:06:07.053 --> 01:06:10.930 align:middle line:84% from my colleague, Charlie Lieberman, who works there. 01:06:10.930 --> 01:06:14.810 align:middle line:84% So these are results of a couple studies that did this. 01:06:14.810 --> 01:06:19.130 align:middle line:84% So they get these cadaver cochlea. 01:06:19.130 --> 01:06:21.390 align:middle line:84% And they know the age at which the person died. 01:06:21.390 --> 01:06:23.390 align:middle line:84% And it can be just anywhere on this range. 01:06:23.390 --> 01:06:25.850 align:middle line:90% So each dot here is a person. 01:06:25.850 --> 01:06:28.970 align:middle line:84% And the graph here is plotting the percent of hair cells 01:06:28.970 --> 01:06:32.770 align:middle line:84% that are present in that individual at death. 01:06:32.770 --> 01:06:36.010 align:middle line:84% And we have IHC-- that's inner hair cells-- 01:06:36.010 --> 01:06:37.650 align:middle line:90% on the left and OHCs-- 01:06:37.650 --> 01:06:39.390 align:middle line:84% that's outer hair cells-- on the right. 01:06:39.390 --> 01:06:42.030 align:middle line:84% And the different colors are just two different studies, 01:06:42.030 --> 01:06:45.200 align:middle line:84% one in 1968 and one from a couple of years ago. 01:06:45.200 --> 01:06:47.380 align:middle line:84% And what these graphs show is that, 01:06:47.380 --> 01:06:51.540 align:middle line:84% as the age of an individual increases, at death, 01:06:51.540 --> 01:06:53.540 align:middle line:90% they have fewer hair cells. 01:06:53.540 --> 01:06:54.940 align:middle line:84% So there's a few people here who, 01:06:54.940 --> 01:06:56.240 align:middle line:90% unfortunately, died very young. 01:06:56.240 --> 01:06:57.948 align:middle line:84% And you can see that they have almost all 01:06:57.948 --> 01:06:59.800 align:middle line:90% of the hair cells remaining. 01:06:59.800 --> 01:07:03.280 align:middle line:84% And then it, on average, gradually drops off with age. 01:07:03.280 --> 01:07:06.700 align:middle line:90% 01:07:06.700 --> 01:07:10.940 align:middle line:84% So this is a graph that plots the hair cell 01:07:10.940 --> 01:07:13.920 align:middle line:84% survival as a function of position on the cochlea. 01:07:13.920 --> 01:07:17.660 align:middle line:84% So the previous graph is just pulling everything together. 01:07:17.660 --> 01:07:19.932 align:middle line:84% And here, we have hair cell survival 01:07:19.932 --> 01:07:21.640 align:middle line:84% as a function of position on the cochlea. 01:07:21.640 --> 01:07:24.020 align:middle line:84% And what this shows is that, for inner hair cells, 01:07:24.020 --> 01:07:27.020 align:middle line:84% there's considerably more hair cell loss 01:07:27.020 --> 01:07:29.540 align:middle line:90% at the base of the cochlea. 01:07:29.540 --> 01:07:34.340 align:middle line:84% And which frequencies get transduced at the base? 01:07:34.340 --> 01:07:36.500 align:middle line:90% High or low? 01:07:36.500 --> 01:07:37.960 align:middle line:90% High, that's right. 01:07:37.960 --> 01:07:39.880 align:middle line:84% So base transduces high frequencies. 01:07:39.880 --> 01:07:43.015 align:middle line:84% So that's consistent, in principle, with the fact 01:07:43.015 --> 01:07:45.640 align:middle line:84% that people tend to lose their hearing at high frequencies more 01:07:45.640 --> 01:07:47.320 align:middle line:90% than low. 01:07:47.320 --> 01:07:51.980 align:middle line:84% You can see that outer hair cells also tend to be lost, 01:07:51.980 --> 01:07:54.460 align:middle line:84% but it's more distributed across the cochlea. 01:07:54.460 --> 01:07:58.640 align:middle line:84% So both the base and the apex have outer hair cell loss 01:07:58.640 --> 01:08:00.840 align:middle line:90% and even in the middle. 01:08:00.840 --> 01:08:04.620 align:middle line:90% So this is a study. 01:08:04.620 --> 01:08:07.660 align:middle line:84% So you can also look at this in lab animals. 01:08:07.660 --> 01:08:12.240 align:middle line:84% And these are graphs that plot five different species of lab 01:08:12.240 --> 01:08:16.399 align:middle line:84% animal-- gerbils, chinchillas, mice, rats, and guinea pigs. 01:08:16.399 --> 01:08:17.880 align:middle line:84% And this is their outer hair cell 01:08:17.880 --> 01:08:22.279 align:middle line:84% survival as a function of position along the cochlea. 01:08:22.279 --> 01:08:24.319 align:middle line:84% And if you compare that to humans, 01:08:24.319 --> 01:08:26.640 align:middle line:84% there's this interesting difference, 01:08:26.640 --> 01:08:29.560 align:middle line:84% that humans seem to have considerably more outer hair 01:08:29.560 --> 01:08:32.612 align:middle line:84% cell loss at the base of the cochlea. 01:08:32.612 --> 01:08:35.279 align:middle line:84% Now, what is it that's different about humans than all these lab 01:08:35.279 --> 01:08:35.779 align:middle line:90% animals? 01:08:35.779 --> 01:08:39.859 align:middle line:84% Well, the lab animals lead very quiet lives. 01:08:39.859 --> 01:08:42.330 align:middle line:84% They sit in a boring room in a lab 01:08:42.330 --> 01:08:46.970 align:middle line:84% and are not exposed to all of the noise and rock 01:08:46.970 --> 01:08:50.870 align:middle line:84% concerts and stuff that normal humans get exposed to. 01:08:50.870 --> 01:08:54.250 align:middle line:84% So you might imagine that this is consistent with the idea 01:08:54.250 --> 01:09:00.490 align:middle line:84% that some component of this is related to noise exposure. 01:09:00.490 --> 01:09:04.229 align:middle line:84% So there's this hypothesis that prolonged exposure to noise, 01:09:04.229 --> 01:09:08.410 align:middle line:84% possibly even that which occurs Incidentally in industrialized 01:09:08.410 --> 01:09:10.930 align:middle line:90% societies, likely plays a role. 01:09:10.930 --> 01:09:12.774 align:middle line:84% And so this gets back to the question 01:09:12.774 --> 01:09:14.649 align:middle line:84% that was asked a little bit earlier, which is 01:09:14.649 --> 01:09:16.350 align:middle line:90% what happens in other places. 01:09:16.350 --> 01:09:19.569 align:middle line:84% So there's a couple of intriguing studies 01:09:19.569 --> 01:09:23.810 align:middle line:84% that have been done that suggest that, in non-industrialized 01:09:23.810 --> 01:09:29.490 align:middle line:84% places, there's quite a bit less of age-related hearing loss. 01:09:29.490 --> 01:09:33.930 align:middle line:84% So this is one study that was done in 1962 in a fairly 01:09:33.930 --> 01:09:36.410 align:middle line:90% remote area in Africa. 01:09:36.410 --> 01:09:38.700 align:middle line:84% So it says, in the description here, 01:09:38.700 --> 01:09:42.160 align:middle line:84% the area chosen is about 650 miles southeast of Khartoum, 01:09:42.160 --> 01:09:44.317 align:middle line:84% the capital of the Republic of the Sudan, 01:09:44.317 --> 01:09:45.859 align:middle line:84% a few miles from the Ethiopian border 01:09:45.859 --> 01:09:47.319 align:middle line:84% and 10 degrees above the equator. 01:09:47.319 --> 01:09:49.800 align:middle line:84% Until 1956, this area was a closed one, 01:09:49.800 --> 01:09:53.880 align:middle line:84% untouched by any foreign culture or civilization. 01:09:53.880 --> 01:09:58.500 align:middle line:84% So they went to this place and measured audiograms. 01:09:58.500 --> 01:09:59.880 align:middle line:90% And the results are shown here. 01:09:59.880 --> 01:10:03.740 align:middle line:84% So this is this group of people called the Mabaans. 01:10:03.740 --> 01:10:07.180 align:middle line:84% And they compared the results to a very large study 01:10:07.180 --> 01:10:10.900 align:middle line:84% of hearing that was conducted at the Wisconsin State 01:10:10.900 --> 01:10:12.720 align:middle line:90% Fair in 1954. 01:10:12.720 --> 01:10:14.720 align:middle line:84% They set up a booth at the Wisconsin State Fair. 01:10:14.720 --> 01:10:17.095 align:middle line:84% And all these people came and had their hearing measured. 01:10:17.095 --> 01:10:20.662 align:middle line:84% And so people in Wisconsin in the '50s show something 01:10:20.662 --> 01:10:23.120 align:middle line:84% that's consistent with the graph that I showed you earlier. 01:10:23.120 --> 01:10:26.260 align:middle line:84% So as a function of age, hearing gets worse, especially 01:10:26.260 --> 01:10:27.540 align:middle line:90% at high frequencies. 01:10:27.540 --> 01:10:30.340 align:middle line:84% So the Mabaans, on the other hand, 01:10:30.340 --> 01:10:34.660 align:middle line:84% show much less of this age-related hearing loss. 01:10:34.660 --> 01:10:35.180 align:middle line:90% Yeah? 01:10:35.180 --> 01:10:38.315 align:middle line:84% STUDENT: What does it mean when [INAUDIBLE] negative? 01:10:38.315 --> 01:10:39.190 align:middle line:90% JOSH MCDERMOTT: Yeah. 01:10:39.190 --> 01:10:41.550 align:middle line:84% I mean, the 0 is just-- it's like a standard. 01:10:41.550 --> 01:10:44.670 align:middle line:84% It's supposed to be like the average person who has 01:10:44.670 --> 01:10:46.610 align:middle line:90% good hearing, normal hearing. 01:10:46.610 --> 01:10:48.750 align:middle line:84% So you can do better than that, yeah. 01:10:48.750 --> 01:10:51.390 align:middle line:90% 01:10:51.390 --> 01:10:54.430 align:middle line:84% So you might think this is circumstantial evidence 01:10:54.430 --> 01:10:57.590 align:middle line:84% that, well, there's presumably some difference 01:10:57.590 --> 01:11:00.212 align:middle line:84% between the people in Wisconsin and the Mabaans 01:11:00.212 --> 01:11:01.170 align:middle line:90% that accounts for this. 01:11:01.170 --> 01:11:02.587 align:middle line:84% And there could be lots of things. 01:11:02.587 --> 01:11:05.750 align:middle line:84% But one candidate is the fact that industrialized society 01:11:05.750 --> 01:11:06.930 align:middle line:90% is very noisy. 01:11:06.930 --> 01:11:09.610 align:middle line:84% We're around noisy machinery all the time. 01:11:09.610 --> 01:11:12.630 align:middle line:84% And that might not be good for our hearing. 01:11:12.630 --> 01:11:14.870 align:middle line:84% So the other study that was done along these lines 01:11:14.870 --> 01:11:17.510 align:middle line:90% was conducted on Easter Island. 01:11:17.510 --> 01:11:21.070 align:middle line:84% This is an island that's off the coast of South America. 01:11:21.070 --> 01:11:23.710 align:middle line:84% It has been called the most isolated inhabited place 01:11:23.710 --> 01:11:25.070 align:middle line:90% in the world. 01:11:25.070 --> 01:11:27.185 align:middle line:90% And I guess, at some point-- 01:11:27.185 --> 01:11:28.810 align:middle line:84% I don't know if this is still the case. 01:11:28.810 --> 01:11:30.935 align:middle line:84% But it was known as the Island of the Great Silence 01:11:30.935 --> 01:11:32.650 align:middle line:90% because it's non-industrialized. 01:11:32.650 --> 01:11:35.233 align:middle line:84% So you don't have modern day machinery there. 01:11:35.233 --> 01:11:37.400 align:middle line:84% At least, you didn't at the time when this was done. 01:11:37.400 --> 01:11:39.120 align:middle line:84% And so what they did is they went 01:11:39.120 --> 01:11:42.940 align:middle line:84% to Easter Island and measured audiograms of people there. 01:11:42.940 --> 01:11:45.840 align:middle line:90% 01:11:45.840 --> 01:11:48.180 align:middle line:84% Now, people who live on Easter Island, I guess, 01:11:48.180 --> 01:11:50.480 align:middle line:84% would occasionally go to mainland South America 01:11:50.480 --> 01:11:52.960 align:middle line:84% for some period of time, like to work for a little bit, 01:11:52.960 --> 01:11:54.340 align:middle line:90% and then sometimes come back. 01:11:54.340 --> 01:11:59.040 align:middle line:84% And so the researchers divided the participants in this study 01:11:59.040 --> 01:12:01.700 align:middle line:84% into those that had never left Easter Island, 01:12:01.700 --> 01:12:04.320 align:middle line:84% and to those that had been away on mainland South 01:12:04.320 --> 01:12:07.040 align:middle line:84% America for three to five years, and those that had been away 01:12:07.040 --> 01:12:08.320 align:middle line:90% more than five years. 01:12:08.320 --> 01:12:10.840 align:middle line:84% And so this is the average audiogram for those three groups 01:12:10.840 --> 01:12:11.840 align:middle line:90% of people. 01:12:11.840 --> 01:12:14.440 align:middle line:84% So what this shows is that the people that 01:12:14.440 --> 01:12:17.040 align:middle line:84% had been on mainland South America for some period of time 01:12:17.040 --> 01:12:20.538 align:middle line:84% had worse hearing than the ones that had never left. 01:12:20.538 --> 01:12:23.080 align:middle line:84% Interestingly, if you take the group of people that had never 01:12:23.080 --> 01:12:25.820 align:middle line:84% left and you subdivide them according to age, 01:12:25.820 --> 01:12:28.600 align:middle line:84% you still see signs of age-related hearing loss. 01:12:28.600 --> 01:12:30.800 align:middle line:84% So the older participants have worse hearing 01:12:30.800 --> 01:12:32.190 align:middle line:90% than the younger participants. 01:12:32.190 --> 01:12:35.010 align:middle line:84% But this, again, is circumstantial evidence 01:12:35.010 --> 01:12:37.690 align:middle line:84% that something about living on mainland South America 01:12:37.690 --> 01:12:40.170 align:middle line:84% causes your hearing to deteriorate. 01:12:40.170 --> 01:12:43.770 align:middle line:90% And it could be noise exposure. 01:12:43.770 --> 01:12:47.690 align:middle line:84% So there's this hypothesis that the incidental noise exposure 01:12:47.690 --> 01:12:51.090 align:middle line:84% that happens with modern life may contribute 01:12:51.090 --> 01:12:52.710 align:middle line:90% to age-related hearing loss. 01:12:52.710 --> 01:12:56.650 align:middle line:84% And I think it's plausible that we evolved in conditions 01:12:56.650 --> 01:12:58.770 align:middle line:90% that were just much less noisy. 01:12:58.770 --> 01:13:00.890 align:middle line:84% And so the ears are just ill-equipped 01:13:00.890 --> 01:13:04.930 align:middle line:84% for the sound levels that we encounter in life today. 01:13:04.930 --> 01:13:07.010 align:middle line:90% That seems possible. 01:13:07.010 --> 01:13:09.730 align:middle line:84% Now, what's much less controversial 01:13:09.730 --> 01:13:13.890 align:middle line:84% is that acute noise exposure can cause hearing loss 01:13:13.890 --> 01:13:15.490 align:middle line:84% through damage to the organ of Corti. 01:13:15.490 --> 01:13:17.530 align:middle line:84% So if you get exposed to something 01:13:17.530 --> 01:13:19.970 align:middle line:90% that's really, really loud-- 01:13:19.970 --> 01:13:23.150 align:middle line:84% it could be gunshots, could be an airplane, 01:13:23.150 --> 01:13:25.455 align:middle line:90% if you're an airplane mechanic-- 01:13:25.455 --> 01:13:27.330 align:middle line:84% we have pretty clear evidence that that's not 01:13:27.330 --> 01:13:28.270 align:middle line:90% good for your hearing. 01:13:28.270 --> 01:13:30.780 align:middle line:84% And some of the clearest evidence 01:13:30.780 --> 01:13:35.060 align:middle line:84% actually comes from this really interesting study of hunters. 01:13:35.060 --> 01:13:38.620 align:middle line:84% So this is a person holding a rifle. 01:13:38.620 --> 01:13:40.420 align:middle line:84% And in this particular study, they 01:13:40.420 --> 01:13:44.640 align:middle line:84% recruited people who, I guess, were aspiring hunters. 01:13:44.640 --> 01:13:46.260 align:middle line:84% And they measured their audiograms 01:13:46.260 --> 01:13:48.420 align:middle line:84% before they learned how to hunt, and then 01:13:48.420 --> 01:13:50.620 align:middle line:84% after they had been hunting for a while. 01:13:50.620 --> 01:13:52.860 align:middle line:84% So the person's holding the rifle-- 01:13:52.860 --> 01:13:56.660 align:middle line:84% and the idea is that there's some-- you fire the gun. 01:13:56.660 --> 01:13:59.180 align:middle line:84% And there's an explosion in here that causes a big sound 01:13:59.180 --> 01:14:00.280 align:middle line:90% wave to come out. 01:14:00.280 --> 01:14:03.420 align:middle line:84% And there's a direct path from the source 01:14:03.420 --> 01:14:05.860 align:middle line:84% of the sound wave and the left ear, 01:14:05.860 --> 01:14:08.780 align:middle line:84% whereas the right ear is shielded from the sound. 01:14:08.780 --> 01:14:13.480 align:middle line:84% And what you can see here are the pre-exposure audiograms. 01:14:13.480 --> 01:14:15.860 align:middle line:90% Those are the solid lines. 01:14:15.860 --> 01:14:19.240 align:middle line:84% So the x's are the left ears and the o's are the right ears. 01:14:19.240 --> 01:14:21.700 align:middle line:84% And then the post-exposure audiograms-- so 01:14:21.700 --> 01:14:23.340 align:middle line:84% you can see that post-exposure-- so 01:14:23.340 --> 01:14:25.940 align:middle line:84% that means after the person started hunting-- the thresholds 01:14:25.940 --> 01:14:26.775 align:middle line:90% are elevated. 01:14:26.775 --> 01:14:28.400 align:middle line:84% But they're substantially more elevated 01:14:28.400 --> 01:14:31.400 align:middle line:84% on the left than the right, consistent with what you would 01:14:31.400 --> 01:14:33.760 align:middle line:90% expect from the noise exposure. 01:14:33.760 --> 01:14:34.400 align:middle line:90% OK. 01:14:34.400 --> 01:14:37.680 align:middle line:84% Eventually, the people who became hunters 01:14:37.680 --> 01:14:40.720 align:middle line:84% passed away and donated their ears to science. 01:14:40.720 --> 01:14:44.520 align:middle line:84% And this is the result of otopathology done on their ears. 01:14:44.520 --> 01:14:47.000 align:middle line:84% So we've got the left ear and the right ear. 01:14:47.000 --> 01:14:49.880 align:middle line:84% And these are the counts of inner hair cells and outer hair 01:14:49.880 --> 01:14:50.420 align:middle line:90% cells. 01:14:50.420 --> 01:14:52.212 align:middle line:84% And so there's three lines, because there's 01:14:52.212 --> 01:14:54.960 align:middle line:84% three rows of outer hair cells typically, 01:14:54.960 --> 01:14:58.320 align:middle line:84% along the length of the basilar membrane. 01:14:58.320 --> 01:15:01.560 align:middle line:84% And you can see that, at the basal end of the basilar 01:15:01.560 --> 01:15:03.160 align:middle line:84% membrane in the left ear, there's 01:15:03.160 --> 01:15:06.400 align:middle line:84% this precipitous drop in the presence of hair cells. 01:15:06.400 --> 01:15:11.340 align:middle line:84% And certainly, the right ear also doesn't look normal. 01:15:11.340 --> 01:15:14.200 align:middle line:84% But it's not nearly as pronounced 01:15:14.200 --> 01:15:15.593 align:middle line:90% as what you see in the left ear. 01:15:15.593 --> 01:15:17.760 align:middle line:84% And we, again, think that's because the right ear is 01:15:17.760 --> 01:15:20.960 align:middle line:84% protected by the acoustic shadow. 01:15:20.960 --> 01:15:24.480 align:middle line:90% Any questions about that? 01:15:24.480 --> 01:15:25.240 align:middle line:90% Yeah? 01:15:25.240 --> 01:15:27.790 align:middle line:84% STUDENT: So have there been studies 01:15:27.790 --> 01:15:30.570 align:middle line:84% where lab animals have been exposed to noise 01:15:30.570 --> 01:15:32.437 align:middle line:90% and their hearing is measured? 01:15:32.437 --> 01:15:34.270 align:middle line:84% JOSH MCDERMOTT: Yeah, there's lots of those. 01:15:34.270 --> 01:15:38.890 align:middle line:84% In fact, that is what we are going to talk about next. 01:15:38.890 --> 01:15:43.970 align:middle line:84% Yeah, so we know quite a lot about the temporary or transient 01:15:43.970 --> 01:15:47.450 align:middle line:84% effects of noise exposure from these studies of lab animals, 01:15:47.450 --> 01:15:49.210 align:middle line:90% yeah. 01:15:49.210 --> 01:15:52.930 align:middle line:90% Any other questions about this? 01:15:52.930 --> 01:15:56.290 align:middle line:84% So OK, before we wrap, if there's 01:15:56.290 --> 01:16:01.290 align:middle line:84% one thing that you take away from this class, wear earplugs. 01:16:01.290 --> 01:16:02.850 align:middle line:90% You're going out to hear music? 01:16:02.850 --> 01:16:05.770 align:middle line:90% Wear earplugs. 01:16:05.770 --> 01:16:09.570 align:middle line:84% You'll be happy when you're older. 01:16:09.570 --> 01:16:11.130 align:middle line:84% And in fact, the thing that I'm going 01:16:11.130 --> 01:16:14.110 align:middle line:84% to tell you next, which will have to wait until next time, 01:16:14.110 --> 01:16:16.570 align:middle line:84% is going to be some particularly sobering evidence 01:16:16.570 --> 01:16:20.550 align:middle line:84% that actually hearing protection is important, 01:16:20.550 --> 01:16:23.620 align:middle line:84% because one of the other things that can happen 01:16:23.620 --> 01:16:27.580 align:middle line:84% is, in addition to these threshold shifts that show up 01:16:27.580 --> 01:16:30.460 align:middle line:84% in the audiogram, so where your thresholds get elevated, 01:16:30.460 --> 01:16:32.738 align:middle line:84% there are also temporary threshold shifts. 01:16:32.738 --> 01:16:34.780 align:middle line:84% So if you get exposed to moderately loud sounds-- 01:16:34.780 --> 01:16:36.540 align:middle line:84% this is like the rock concert effect. 01:16:36.540 --> 01:16:38.680 align:middle line:84% You go out to a nightclub or a concert. 01:16:38.680 --> 01:16:40.100 align:middle line:84% You probably have this experience 01:16:40.100 --> 01:16:43.620 align:middle line:84% that you leave and you can tell that your hearing is not normal. 01:16:43.620 --> 01:16:44.680 align:middle line:90% It's harder to hear. 01:16:44.680 --> 01:16:46.700 align:middle line:90% Maybe you have some ringing. 01:16:46.700 --> 01:16:48.040 align:middle line:90% And that's measurable. 01:16:48.040 --> 01:16:50.180 align:middle line:84% So these are temporary threshold shifts. 01:16:50.180 --> 01:16:52.180 align:middle line:84% So if you went and measured your audiogram right 01:16:52.180 --> 01:16:54.138 align:middle line:84% after you went to the nightclub, the thresholds 01:16:54.138 --> 01:16:55.300 align:middle line:90% would be elevated. 01:16:55.300 --> 01:16:56.680 align:middle line:90% But then they go back to normal. 01:16:56.680 --> 01:16:59.380 align:middle line:84% You wait a day or even a few hours 01:16:59.380 --> 01:17:00.780 align:middle line:90% and they go back to normal. 01:17:00.780 --> 01:17:03.460 align:middle line:84% But as we will see next time, there's 01:17:03.460 --> 01:17:06.820 align:middle line:84% reason to believe that, even though your thresholds return 01:17:06.820 --> 01:17:09.460 align:middle line:84% to normal, your auditory system has potentially 01:17:09.460 --> 01:17:10.860 align:middle line:90% been altered by this. 01:17:10.860 --> 01:17:15.570 align:middle line:84% And we will hear about that next time. 01:17:15.570 --> 01:17:23.000 align:middle line:90%