Amar Bose: 6.312 Lecture 25

AMAR G. BOSE: Today, we have the remainder of the list of last time. I'll go down these, maybe in that order. And if you have questions, as usual, at any point, ask. And there may be things that some of these topics bring up that relate to something you want to know. I don't mind going off the exact list here.

Let's talk a little bit about localization of sound. Horizontal-- by the way, all this again has to do, keep in mind that picture we do with the three sets, physical, physics, and psychophysics. And all that we're going to-- when I'm talking psychoacoustics here, I'm just peeking through the door of it. There's a whole world of information there. I'm peeking through the door just to show you what it's about, to show you that you've got to be cognizant of the end user when you design a product, any kind of a product. And I'm relating some of the factors in psychoacoustics that are important to sound reproduction.

Now, localization is one of them. And we briefly alluded to this earlier. If you're discussing horizontal localization, you use two mechanisms. And the crossover between the two mechanisms occurs at about 1 and 1/2 kilohertz.

Below that, you use time delay. Because below one and a half, the head is about, the way from here to here is about a wavelength at 1 and 1/2 kilohertz. Below 1 and 1/2 kilohertz, the wave goes around here without much attenuation.

So a sound over there gets here and gets here, the same attenuation, but small delays, very small delays. You configure out, by the way, what angles-- you can perceive very small angles. And that gets down into measurements in tens of microseconds sometimes.

But you learn this. There are people who even here say you don't, you're born with it, and all this and that. But if you look at a baby, for example. It hasn't heard any sounds. It's recently born. And you speak to the baby, it doesn't know where the sound's coming from.

But then it correlates the experience it had with what it saw. And it soon learns that whatever came to them before that actually was over there, they recognize that sound. And then they don't need to see. They know what is over there.

Just like when they go to reach for something, when they go to reach for a glass, the arm goes in all sorts of ways. And they use feedback. That's the whole process of our learning. And they don't do any more of what caused the arm to go that way. They do more of what caused it to go this way.

They come to the glass, knock it over. And then the next time, they're able to slow down a little bit. So they're always measuring error between what they got and what they wanted, and then correcting. And that's absolute feedback loop. And you do the same thing in the learning process throughout.

Now in any case, once you've gotten there below 1 and 1/2 kilohertz, you use time delay between the two ears. And this is very easy to show because all that you need do is take a person into an anechoic chamber, put on a set of headsets, and then put a pole out from the head and put a microphone here. And do another one here. And then feed this microphone to this ear and this one to this ear and put a sheet in front of the person.

It is all anechoic because you're trying to find directions, and you don't want sound coming from all over. And then you basically have the person here, and you have a sheet here in front of them. And then you have sources that are arrayed horizontally behind the sheet. And you have numbers on the sheet, for example. At least, that's the way we used to do it here.

And you can move the source by connecting this loudspeaker instead of this one, or whatever you want. And you can ask the person, then, where is the sound. And of course, with these ears, the time delay is much larger for a given angle than it is for these ears. And so the person will take a sound source that's moved from here literally to here and say it went over there.

And you just calculate the difference in the distance between his real eras and his new ears, and you that, in fact, it is delayed, that he's paying attention to, that he's cuing on. Of course, the person doesn't know that. But that's actually what's happening.

Well, above 1 and 1/2 kilohertz, as you know, an object, the sound doesn't get around it. It gets attenuated as you go around. And so if you're trying even in measurements to get the time delay between two signals, one of which is very small and one of which is very big, it's pretty difficult. And so you've learned to change, and you notice that it's louder in this year than it is in this ear if it's coming from over there. And that's the mechanism that you use above.

Now so much for the two mechanisms that describe the horizontal. What about vertical? How do you think that works?

Well, how many would say that you think you have vertical localization? 15%, 20%. How many would say you think you don't? 10%, maybe. Huh.

Well, you have a type-- let me just see. Yeah, good. Big set? Yeah, this is good enough. Now I have to take somebody that is not you, for reasons that I will tell you afterwards.

Here we go. I'm going to ask you to-- not yet-- I'm going to ask you to close your eyes. And I'm going to jiggle the keys around somewhere. And you point to it. But don't open your eyes at that point.

And then I'm going to maybe move the keys somewhere. And I'm going to do it twice. And then you point to it-- in other words, you're going to point twice.

But keep your eyes closed until it's all over. We'll tell you what happened. OK. Here, close them now.

[JINGLING KEYS]

AMAR G. BOSE: OK. what happened was when they were down here, you pointed about up here. When they were up here, you pointed about up there. You were always about 15 degrees or so above. Now just do one more experiment like this, twice. I'll do it twice. Close your eyes, please.

[JINGLING KEYS]

AMAR G. BOSE: Thanks. I did it three times. Very good. Horizontally, you did very well. Vertically, you were off, but off by the same amount.

Now, I picked you because you have a nice, symmetrical head. Which is important because when sound's in front of you, there's no time difference between the two ears, right, if it's in the vertical plane. So basically, time difference is out of the question. You can't use it. It doesn't exist for vertical localization.

The only thing that you can count on is that the spectrum changes. In other words, when the sound is in the front, the spectrum is whatever happened when you tried to go around the head. At low frequencies, nothing happened. And at high frequencies, there was some loss as you went around here, as we know, when the wavelength gets small.

Now when it's coming from up here, there's a different dimension. And so the highs lose more. The lows are the same.

The highs lose more coming from here, around this part of the head, than they do from here. There's another reason. Well, I'll get to that.

So what this tells you, and this is why I took the keys from one person and used another person for the experiment. If I'd used the person that owned the keys, he's accustomed to their sound. And he probably would have been able to localize vertically right on it.

You have relative, but not absolute, in the vertical plane. Because it's only a spectral change. And so if you have never heard the sound before, what happens is if you've never heard it at all, if it weren't a set of keys, if it was just an artificially generated sound, and you take a person into an anechoic chamber, and you have speakers like this, and you make some kind of a noise through the speaker, and you ask him where is it coming from, if it's foreign to him, he'll always point right here.

Then you tell him-- forget it, this is another experiment. This is the one I want to do. Then you change the spectrum of the noise by the difference in the frequency response measured this way and that way. You put that in as a filter into this loudspeaker. And you can make this person then say, oh, you moved the source up.

In other words, I changed the spectrum, just like we did on the keys. The spectrum changed to his ears. So we change it electrically in here, and the person says up, it's here.

Now while he's saying this, you bring another person into the room. You sit him down. You say, where is the sound? He'll say, here. He's never heard it before. He has no absolute vertical localization.

You take the filter out, the first person says it returned here. The second person says, oh, it went down here. So all of these things that you learn about the behavior of the person, you use in the design of your equipment.

In acoustics, for example, I talk to so many groups, I'm not sure what I've told each of you. But when we first went, at the company, into the automobile OEM business, I gave a talk about localization to a particular car company and told them that the loudspeakers should be on the floor. By the way, that isn't on the list. And if you want to know why they should be on the floor later, raise your hand.

But I told them that the loudspeakers should be on the floor. And they're always up in the instrument deck. They always were until that time. And told the reasons why.

And told that nobody would know what. You can set the right spectrum, and nobody will be able to know that they're on the floor. Don't worry about it.

And being there in a particular place has an enormous other advantage. And so everything seemed to go all right, not too many questions. A little bit of frowning I could see in the audience.

And about three days later, some of the engineers of that company called engineers of our company and said, "Dr. Bose be damned," exact words, quote. "If loudspeakers were meant to go on the floor, God would have built your ears into your ankles." Exact statement. Turned it into a religious argument, of which there was no winning.

So basically, if you know some of these things about perception, it gives you freedom to do other things. And I'll give you-- this wasn't planned, but here we go. I'll tell you, even though you didn't ask why those speakers were put on the floor.

Turns out that if you have a driver and a passenger, and you have the usual loudspeakers up on the-- this is the windshield. You have them up here. This thing, you can even see it in the slanted windshields on a sunny day, you can see the speaker reflecting there.

Well, that's the image. You're getting it. And you're looking right into it.

So this one comes to this fellow. This one comes to here. This is left channel, right channel, the way it's in many cars still today. Now your balance control is set in the middle.

Guess what this fellow hears? This thing is so close to him, and this is so far away. Plus the polar pattern is something like that at the high frequencies.

And so this fellow is saturated with left, this one saturated with right. So the driver says, ah, I gotta fix this. He'll crank up the right one and crank down the lower one.

So this fellow now gets blasted totally out. And he gets here. And with a little bit of luck, this much of an angle, if he blasted enough, this much will come over. And he'll have a balanced sound.

So it turns out that you can beat this. If you put the speakers-- I'll tell you why they're in the floor-- but if you put them on the floor instead. Or you put them on, let's say the sides, but over here. Imagine the speaker as a flashlight at the high frequencies. At low frequencies, it's a light bulb. So it has a pattern, let's say, like this. And what the pattern is is very important when you do this, according to the dimensions.

Now this fellow is pretty well off-axis of this one. This is the left. This is the right. As is this fellow. Everything's symmetrical.

But this fellow is much more on-axis here. See, here's the relative strengths. This person over here is getting much more strength from that flashlight beam, if you want to think of it, than this one. So by doing this, getting the polar pattern right, such that the ratio of this strength to this is just right, you find out you can actually have both people perceiving balanced left and right channels sitting there. And then-- yes?

SPEAKER 1: --the same thing, putting it up in the dash, but angling them so that the direct pattern reflects off the center of the windshield and goes to the opposite side?

AMAR G. BOSE: Yeah, it might be possible. We've not yet been able to do it because-- or you've not been able to do it. You can't get it far enough away from the windshield to do what you want yet. So you wind up with your mirror image right on top of the thing.

Basically-- now I forgot what I was going to say. Oh, yeah. I've just thought about on the last day, when we meet on Friday next week, I'm going to give you a couple of quotes that I think will be meaningful to you later on, from two Nobel laureates. And one I already mentioned.

But here's something where you make an advance. You actually can get the darn thing, the sound balanced for both people, which you could not do in this situation. You can't electronically achieve it.

Here, once you've achieved it acoustically, if you have a balance control, it can do nothing but foul it up. And there was no way that we could get any of the car companies to take away the balance control. Oh my god, we'd have an inferior radio. Every other person, every other car company has a balance control. If you don't have one, you don't have a viable stereo product. Oh my god.

And what happens? You go through the car wash. And somebody gets in to clean the interior of the car, turns the controls. The person drives for months. It's happened to me.

God, something's wrong here. It's terrible. Oh my God, the thing. You don't realize it, but you drive along, and you realize something's not right, finally. But you've lived with it for a while before you sort of sense it.

So when you make progress, there are always a chain of people pulling back that say, oh, we can't let off this knob. Because it's always been there, and people won't buy it. They won't buy the car, for god's sake, if the thing doesn't have a balance control on it.

So as you make progress in your careers, remember that there will always be people pulling at your tail to get you back into conventional things. So again, a little knowledge of how localization-- oh, on the floor. On the floor because for the polar patterns that we could get, we needed the distance. And if I'm the driver of the car, there's more distance if I go to where the front door opens, just down in that corner. There's more distance then if I were to put it in here, in the upper part.

Plus, if it's on the floor you have the intersection of two surfaces. And remember what happens in each? You get a mirror image.

So the low frequencies, which are the hardest to radiate, which require the most excursion to radiate, you get a boost from that. And as long as you are designing, you know the whole acoustics of the car, then you can equalize for all of this. And you have a better dynamic range that way, also. So there are all sorts of good reasons for that.

And yet, it was like pulling teeth. There were window motors in the car, things that wind the window up and down. They had to be moved.

And car companies, I can remember one European company told us, look, it would cost us $2 million to retool the door. We can't do it. You're going to put the speakers up here. And we said, thanks, but no thanks because it'll ruin your name and ours. And the next week, they came, and they spent the $2 million, not with us. Yeah?

SPEAKER 2: Aren't the polar patterns frequency-dependent?

AMAR G. BOSE: Yes. Absolutely. And what you gotta do is work on the best average you can at the high frequencies. So OK, that's an example of where you can really achieve something better just by knowing what the perception is. I gave you the example of the color television. Now, let's talk a little bit about other things that can also be important in design. Yeah?

SPEAKER 3: [INAUDIBLE]?

AMAR G. BOSE: OK. Let me see. Yeah, that's where it belongs. Thank you.

Carlos is reminding me of the cocktail party effect. There are many papers in the '60s, respectable research in a good journal, the Journal of the Acoustical Society of America, entitled "Cocktail Party Effect." And what the cocktail party effect is, is related to the fact that if you go to a cocktail party, one of these things in which most of the lips are flapping before the head gets in gear, especially after people have been drinking, basically you're talking to somebody right in front of you.

And there's a story over here that seems very interesting. And you want to make polite conversation, but that thing's attracting you. And so it turns out that you can focus on that.

Anyway, I did talk about depth perception, but I might say a couple words here to tie it all up after I'm done. What I'm talking about now has nothing to do with it. So you can focus pretty well on that, get the meat of it.

And then the question comes, people ask, well how can you do that? And at first, people thought, well, gee, that was a real granny story. And you were excited. And you could mentally focus on what was interesting to you.

Well, then you go along. The next part in, you wear a very expensive, most expensive microphone. And the same thing happens.

And you said, now I've got it. I can get all the gory details because I've got it on the tape recorder. You come home, and it's not there. Hmm.

So that was the subject of almost a decade of work. What is going on? It turns out that you can actually focus your localization.

My own thinking, and I'm not researching in that field at all, so consult others before you make any conclusions on what I say. But the experiments that I have made indicated something like to me, and that's at a low level, that it's very much like a cross-correlation process. And you're able to control the delay of that cross-correlation, which would optimize your response to that direction if that's what you wanted.

And basically, it's known that you can do it. And if you put a binaural-- in other words, if you went to the cocktail party with two microphones right here, little ones, and then you recorded each track, and then you played it back to headphones, then you can do it again. The information would be there. But if you once recorded it in a single microphone, it's all gone.

So there was a lot of research from the people that are interested in what goes on in the auditory process as to what was enabling you to do this. Was it after all this went to the auditory cortex, after all the signals went there, or was it before? Were you able to optimize on the polar pattern?

OK, Fletcher-Munson, this is work-- some of the best work in both physical acoustics and psychoacoustics in this country was done in Bell Telephone Laboratories before they were broken up. And a lot of it dates back a long time. The Fletcher-Munson curves date back to about 1934, I think. They have to do with the threshold of hearing and the sensitivity of hearing as a function of frequency. They look something like this.

Frequency here, and up on this axis, I'm going to put SPL. This is the SPL applied, whether you do it one ear or two ears, it's applied to the ears, pure tone. And the vertical axis will tell us how much you have to apply to get the person to just perceive it.

The way this is conventionally measured is they will start with a very low tone. The person will be wearing a headset in hopefully a quiet environment. They will raise the level until the person, he's told to press the button when he hears the tone. And then they will lower the level again and then raise it.

And you can soon find out where it really is, you got a band, and where it is that the person can hear it. And you get a curve for the threshold of hearing that looks about like this. And this is called often the reference pressure in acoustics. And that is 0 dB, we call on this scale the .0002 microbars or something.

I think that's the same one as the 2 times 10 to the minus 5 newtons per meter squared. I'm not sure. That's 0 dB. Now the interesting thing is, that occurs at about somewhere where they have about the quarter wavelength of the ear canal maximum pressure, 3.5 kilohertz.

Now if you go about down here to 30 hertz, lo and behold, you find this is 60 dB. That's amazing. 60 dB, 1000 times the signal, the sound pressure, you need at 30 hertz Hertz so that the person can just hear it in a quiet environment. 1000 times what you need at this frequency. So at each frequency, then, this is the curve that you need to just hear the sound in a quiet environment. We'll talk about that.

Now that's one type of measurement. The next type of measurement is more difficult to make, but they did. And that's they raised the signal level at all points by 10 dB. And they got another curve, and it looked something like this, let's say.

And keep on doing that until we get a curve which is almost flat. And a that point, the experiment stops because that's the threshold of paint. This is about 120 dB. This might be 10, the applied signal.

And so this curve gets flat as you go up in intensity and gets very bowl-shaped as you go down. And that's a property, that's just a terminal observation on the ear. You can go in and make acoustic or mechanical or electrical models of the whole process and get a lot of information as to why. But a terminal look looks something like this.

Now this is called Fletcher-Munson after the two fellows at Bell Labs who did it. And by the way, an interesting point, when you do research on perception, you might think, oh, boy, this is a terrible thing to do, 2000 people and whatnot. Well, if you have to do 2000 people to get the measurement, you're in bad shape.

You don't want to look at that measurement. Because if the variance is so bad, if you do 10 people, and the variance is all over the map, the measurement isn't going to be very useful to you. The end result, the knowledge, because it's almost like random.

So it turns out, if I remember correctly, Bell, Fletcher and Munson did this on 13 people. They took some elementary precautions like putting a flashlight here and making sure the light didn't come out over here. But that the people had hearing, somewhat normally, you might say. But once that's established, and you can establish those kind of things pretty easily. Once that's all established, 13 people, and this thing has held since the '30s.

Others have repeated it. There are curves made in England. There are another set of curves, somebody else, I've forgotten the names of them. But they're all similar.

And we've done it here. In fact, in doing it here, I can remember, we couldn't even use the anechoic chamber because the anechoic chamber has a low-frequency vibration in the thing. And when trucks go by on Vassar Street, that sets that whole thing into bouncing around at 20-something hertz. And you can't use it.

And so we had to go into the best place we could find. Because to get this curve, the best place that we were able to find is in one of the rooms in the basement of Kresge Auditorium in the middle of the night, and getting the maintenance people to shut off all the air conditioning and all motors and heaters of any sort. Then you could take this curve.

These curves are harder to take because now what you've got to do-- how do you find out what his perception is when he can already hear it? You go along, these are called equal loudness contours, the other ones, the ones that are these. You go along, and you give him, let's say a 1,000, 1-kilohertz tone of the given amplitude. It's 10 dB up from what you had, let's say, on this one.

And then you go along over here, and you give him a tone. And you say, when do you perceive this loudness to be the same as this? You go back and forth between the two tones.

And that's harder to do. You're asking the person to take something at 60 hertz and match its loudness to a kilohertz. But that's what you do.

And it turns out you can do it. The people, after you do 20 minutes of it, my experience is they come out totally perspiring. it's a real concentrated effort to do it. But it can be done.

Now, engineers have taken this curve and said, aha. When music is played back, if music in the concert hall might get up into this region here, 105, 110 dB. And then it comes back in the room, and it might be played at 60, 70, 80 dB. See the difference now. The frequency response of the ear falls off.

If you turn upside down this curve, you get the frequency response of the ear. Because this curve the way it's written says you need more signal to hear the same thing when you go up here than you did here. So turn it upside down, and it's like a filter characteristic. So it says the ear falls off.

Well, a lot of the people, a lot of the companies mistake this. And they see this bowl-shaped thing over here, and they say, oh, it falls off at high and low frequencies. And hi-fi magazines are full of that kind of stuff, too. It doesn't.

It does, but it doesn't relatively. They stay relatively constant, the distance between them. Whereas all these pairs on the low end get squished together. And so you don't have a relative frequency problem at the high end, when you go up and down.

But in any case, what came out of this was when the hi-fi enthusiasts and companies got ahold of it, they made the inverse filter. Which the inverse to your filter, which is this one, OK? So what they do is they start off at-- it's called a loudness control. You've seen those things.

You start off flat when it's full up. And then as you turn the volume down, you begin to get boosts here, and mistakingly, boosts here. And then you go down further and further, you get a steeper boost.

Now the interesting thing is, and we alluded to that very briefly before, all of the systems that I've ever seen always have an on/off switch for the loudness control. And it's there for a reason. Because it makes speech sound horrible. And it makes music sound acceptable.

And I mentioned this in between some other things before and said we'd come back to it later on. I don't know how much of it stuck. Do you want me to explain again why? How many want me to explain again why? Maybe a third, very briefly.

Very briefly, it's simply because you've never heard music composed at a different level, unless you listen to radios, live performance, you've never heard music played at a level different than what the composer basically designed it for. And he put in it the balance of his tones and notes, and he knew what instruments he was composing for, so that it would sound right at the level that he expected to have it in the halls of the day. But you never heard a loud sound at 40 dB less than that, or 30 dB, or 20 dB less, a live sound.

So when you hear it now coming out of your radio in your house, you miss the things that you heard live, the bass and whatnot. And so if it's cranked up appropriately, it brings you back to a type of reality that you never had and you could never-- well, I suppose I can't say you could never get. You could bore a hole in the wall of Symphony Hall or something and stand outside.

But basically, it's a sensation you weren't meant to have. And so you restore something that's totally artificial now. But it can move you toward believing that it's more real.

But you can't do that with a voice because you have heard everybody's voice speaking softly, speaking loudly. And when you hear the person speaking softly, this filter was operating for you. When you hear the person speaking more loudly or shouting, this filter was operating. And you know it. That's what has been reality to you.

That's why I mentioned before that if you take a two people that know each other well and put headphones on one, put them in one room-- sorry, put a microphone on this fellow, put a headphone on the other one. And tell the one outside now, adjust the sound 'til its most realistic, he will always set the volume so the sound pressure is what he would hear if he were standing right next to you. If you talk softly, he'll never crank it up out there. You watch what he does, and he brings it to soft. Because he's used to the fact that the spectrum of your voice changes.

And so if you artificially, then, boost this thing up with a big boost like this, and a voice comes along, oh my god. We already saw the hormone effect with the cardioid microphone. And now you get, on top of that, this loudness control. And the whole thing sounds like it's in a big box, like a jukebox or something. So that's why they have the off switch.

But without the research, they don't even know why they have it. They know why they had it, because it doesn't sound good sometimes. But again, you must, in whatever discipline you go into, study the end user and see what that person perceives and what he doesn't. Yeah, yeah. OK.

[? SPEAKER 4: --sort of like ?] A-weighting, this curve?

AMAR G. BOSE: A-weighting, I'm not going to discuss it now. But all the different weightings are noise kind of weightings that give you a rating, supposedly, of how if you put the sound through a filter, a weighting is just a frequency curve. And you put it through the filter, and you rate different kinds of sounds and noises. And really, they are today, I would say, honestly, inappropriate specifications. We bump into this, for example, in the active noise cancellation, which you people will experience.

By the way, I don't know how we're going to do all this. There's going to be over 150 people going through this. Because I forgot that the course is also being taught by last year's head TA out at the company for engineers out there and for WPI. And so they're all coming. But somehow, I think you're going to get to this.

But when you do the headsets, active noise cancellation, it turns out that you can dramatically reduce the noise, like 25 dB. But on the different weightings, usually the A-weighting, you don't get a reduction. And so the specifications of the government, for example, for their procurement for headsets, are using these weightings that were thought of for various purposes a long, long time ago. And they're obliged to meet those things.

And here you come along with something, when you put it on a person, they say, oh my god, what a difference. And the meter says, essentially, trivial difference. But the contract is governed by the meter. So these are all the problems, again, of making progress. But you've got to fight with them. So yes, it was another question you had. Yeah.

SPEAKER 5: These curves you showed were all for single tones, right?

AMAR G. BOSE: Yeah.

SPEAKER 5: If you took a bunch of tones that were all individually below the threshold level, so that the total energy would be above 0 dB SPL, would you be able to hear it? Or is it purely individual frequency [INAUDIBLE]?

AMAR G. BOSE: OK. That get us a little bit-- if they were separated, that's another issue which I wasn't going to discuss called critical bandwidth. If they're separated enough, they'll be like they're independent. What happens, I'll mention critical bandwidth a little bit.

The cochlea of the ear, which if you unwind the thing-- you can't, it's made of bone. But if you did, it's about 3 and 1/2 centimeters long or so. And it would be sort of looking like this. It has a separation down here, basilar membrane, and it's fluid-filled.

And what happens, this membrane, which is actually narrower, interestingly enough, at the open end. This is the oval window inside the ear. Ear canal comes in here, and then the middle ear, with the various bones, anvil, et cetera, hammer. They get this window going, and the window gets the fluid going. And that gets the basilar membrane moving.

Now it turns out that this thing acts just like-- and that's important to realize-- it acts like a cone filter, if you wish. A pretty broad one, but if you put a low frequencies in, you excite the membrane down at this end. And if you put higher frequencies, it's here and here and here for higher and higher frequencies. And it didn't matter if you changed the phase of that thing. It's measuring the energy it gets going.

In other words, if I change the phase of the signal, I'm still delivering the same energy in here. And that's one of the reasons that monaural phase, you can do all these things like move your head when I'm speaking, and I still sound like me, with all the millions and millions of phase shifts in normal modes you're going through. It's why you could put the all-pass network that we talked about in a system and come out the other end with a horrible square wave response and still, because all you do is change this phase, and come out the other end with something that sounded the same. Now what was your question again? Because I don't know why I was talking about this.

SPEAKER 5: The question was, if you put the same kind of total energy in, but it was all below this threshold--

AMAR G. BOSE: Yeah, yeah, yeah, yeah. OK. Yeah. Now, it turns out that there's a thing called critical bandwidth. Namely-- well, before I can introduce that, now I've got to go to masking.

If you put a tone in that you could hear, let's suppose I put a tone here that's a dB or so above the threshold. This thing might be, I don't know, 500 hertz or something like that. Now I put another tone here that's well above the threshold. It turns out that I can't hear this one anymore. And that's what's called masking.

In other words, when you put a tone in here, what happens is there's a masking curve. It looks something like this. It's bigger slope on this side then it is on this side. And this curve says that at this frequency now, the new threshold of hearing would have to go up to here before you could hear it.

In other words, let's suppose you put this thing in at 80 dB or something. Maybe the masking tone was 80 dB, let's say. Then you might have a curve which came up as high as 60 dB here. And so out here, this thing that you could hear at 1 dB above the reference level before, now you have to raise it to 60 dB-- well, 50, maybe, or something-- to be able to hear it.

So the interaction of these tones is a very interesting thing. And I'll get to that in one second. This has to do, by the way, it is a very interesting consequence of that. Namely, it turns out that if you make experiments with what bandwidth you need to reproduce so-called audio musical signals, if you come along here with a filter that's really a brick wall, it goes right down like that, that's the transfer function of the filter.

What frequency-- let's say I start out here with one that's very broadband at 100 kilohertz, and I bring this thing down, down, down to cutoff frequency. Where would you think, and this is not for people of my age. It's people that are 18 or so. Where would you think that you could bring this thing down in frequency before you began hearing it, and that something happened to the music reproduction, assuming you could have this marvelous wide band system? Anybody want to guess how far down in frequency this filter could come before you would notice? Take a symphony orchestra with the full range of instruments.

SPEAKER 6: 16 kilohertz?

AMAR G. BOSE: 16 kilohertz, one guess.

SPEAKER 7: 10.

AMAR G. BOSE: 10?

SPEAKER 8: 20.

AMAR G. BOSE: It's really interesting because whatever audio file nature you had in is now washed out. Because I was expecting big numbers like 20 kilohertz. Everybody advertises 20 to 20 kilohertz. It turns out that you're very close. 15.

Now people will tell you, oh, that's nonsense. I tried it at 15, and I could hear the A-B difference. Yeah, they tried it at 15 with a filter that did this. And they were hearing this, the difference in the music.

You really have to have a sharp filter to be able to determine that number. You've got to go down at least 36 dB per octave on the filter to make the experiment. But it turns out to be that.

By the way, that reminds me of when we were doing the early research into sound, everybody had published this absolute necessity of 20 to 20 kilohertz. And in these days, late '50s, they all had loudspeakers on the market, and they claimed that. And so we decided, we made some measurements of ones in our anechoic chamber here, and none of them made it.

But we decided to make one, since the numbers were there. And this was just the beginning. We had no idea what was necessary and what wasn't.

And so we made one that actually went down. We electronically equalized it. It wouldn't play at huge volume levels, but it went down to 20 hertz.

And there was a station on the air at the time, and it was WXHR. And it was all classical. And we were doing these experiments in the middle of the night. And it turned out that every other record that they played, the cones on the loudspeaker, you couldn't hear anything.

But this was the eighth of a sphere that you could put in a corner with 22 of the same one we used for the impulse response. But every other one, if you looked at the loudspeaker, the cones were going like this. Huge excursion, just going like that. And it wasn't making any sound. Then the next record they played, it didn't do it. Next record, yes.

So we called them up. And we said look, you fellows have an interesting problem. You have two different sources that you must be playing. And every other one has this really low-frequency noise in it.

And they were shocked. They said, how do you know this? Who are you? They said, yes, we have a turntable with rumble in it. But nobody had ever complained.

Now, if, in fact people really had loudspeakers in their home, and there were thousands of them in the area that did, that bought all sorts of expensive stuff at the time that went down to 20 hertz. What would have happened if they did as advertised, the cone motion would have been so enormous from a 10-inch cone, or a woofer, that it would have run out of the linear range and, at 20 hertz, would have distorted the whole spectrum of music that came through.

So from that experiment, when they told us they never had this complaint at all, then we knew right away that the speakers out there never went down there. And we didn't have to survey all of them on the market. Because somebody would have bitterly complained about that.

It didn't make any sound because even motion like that at 20 hertz of a 10-inch cone isn't going to produce anything that you're going to hear. The curve is too high up here. But it sure used up the linear region and screwed up the rest of the performance. So OK.

Oh, by the way, when we introduced the first product, the 901, it was the first product of the company, we sent it to Stereo Review it was at the time. And we said the specs were 30 hertz to, I think, 15. And they came back to us. And they really were because we tried very hard to cut it off at 30 hertz because all the record players would just produce something you wouldn't hear. But you would just use up the linear range.

And they said, oh my god, you'll never be able to sell a single unit if you say 30 hertz. Everybody says 20. So we said, well, I'm sorry, it's 30.

And so they were trying to be kind to us. And when the thing was published it said 20 to 20,000. Unbelievable. Again, the tie to the past. You gotta do what everybody else does.

And always remember one thing. If you do what others do, be it advertising or anything else, you will never win. A person who follows somebody in a race never has a chance of winning the race. He's got to go on a different path than the person in front of him in order to pass him.

So that may sound like motherhood and apple pie now, but when you get out in industry, and you start going on the other path, you're going to meet opposition. Just realise that you're not alone. Everybody that's made progress, you're in the same group, and it's a nice group. OK. There was, yeah.

SPEAKER 9: About that masking, I would think that if you had one tone that was louder, and instead of having another tone that's very close, it seems like it would do a good job masking the same note maybe three octaves above or whatever. Also, it might have to do with the musical octaves.

AMAR G. BOSE: Yeah, no, it doesn't. The curve has a much lower slope this way than this. So lower tones mask higher tones much more than they mask tones that-- here's the masking tone at this frequency. And that will mask all of these fellows way out here much more. Over here, they're not even touched at the same distance out.

SPEAKER 9: --up? Because if there were two tones right next to each other, wouldn't you hear a beating?

AMAR G. BOSE: Oh, that's another whole phenomenon. When the two tones get very close to each other, your ear isn't linear. And you generate a beat product, which you hear. And when I say your ear, I'm talking down all the way through to the audio cortex, basically.

And in fact, that was used as an argument in Phillips, one time, in Eindhoven in Holland, they developed-- what they did, you had a homework problem about a copper band around a voice coil? You didn't? This time, we didn't do it?

OK. If you wind a voice coil around a copper band, what happens is it cancels the voice coil inductance. But it doesn't enter into the rest of the equations. So you get a six-- remember, the voice coil inductor in the speaker model was right in the beginning. And if you could tell that inductor, you'd get a 6 dB per octave gain at the whole high end.

And so they had done this. They were, I think, the first one to produce a loudspeaker would like that. And I went to see them. And this was a long time ago, about 1960. And they were touting this thing, and how great it is.

And I said, well, what use is it? And they said, oh, it's very important because harmonics of the violin and different instruments exist at 22 and 23 kilohertz. You don't hear that, but you hear the beat, which is at 1 kilohertz.

Now it turns out that the model of the year, the linear stuff is all dead. You go through the linear filter inside before you ever get to any of the processing that you might consider nonlinear. And so that's impossible to hear.

And so continually, for about two or three hours of the demonstrations that were in that place, I asked them, well, can I hear it? And, oh, yeah, yeah, yeah. But it never was offered. It's inaudible.

So that's the other example in engineering. Don't get yourself-- when you have a technology, and it doesn't map into the space of perception, doesn't map into a different point, don't sell it. Don't try to keep touting it. There's too much of that in the world, where things that you can measure but are useless in terms of perception.

SPEAKER 10: Quick question about the masking stuff. If you have a tone, assume that you do. I guess, does that relate to if you have something, if you're playing a signal, and you have a signal below it that has less energy, if one signal's masking the other, can you still sense that-- there a way to sense that second signal? This sort of relates into--

AMAR G. BOSE: You're less conscious of the second signal, of the one on the high end. That, to a certain extent, has to do with this also, this limitation. In other words, there's no musical instruments with their fundamentals up here.

There are harmonics. And most of their energy is way down here. And so even though you might have some sounds in this region, they're well-masked by this, by what's in there.