.The complete transcript turned out to be much more space-consuming than we had anticipated, so we are reluctantly parceling out the rest of it in short installments to make room for more up-to-date matters.

The first two installments of our seminar have created a veritable fan club among our readers; this group of audiophiles would like the transcript to go on forever, and they apparently go to it first, before anything else in our pages.

Others have told us that they want equipment reviews, not all this loosey-goosey rapping. To us, the decisive factor is always the possibility of increased knowledge, regardless of the editorial framework, but we concede that after two issues our mainstream concerns should no longer take a backseat.

To understand and enjoy to the fullest these shorter Parts III, IV, etc. of the seminar, make sure you have read Parts I and II, including the capsule bios of the participants.

(Here we continue exactly where we left off;

in the middle of a discussion of binaural sensitivity to very small time offsets, on page 51 of Issue No. 14.)

LIPSHITZ: So there's no contradiction in that statement. And I think the point is, once the rise time of the system-or, said another way, once the bandwidth of the system-sufficiently exceeds the rise time or bandwidth of the hearing system, there is no point in having it any faster because it doesn't substantially or significantly alter what gets through. It's rather like, once the signal-to-noise ratio of your recording sys tem is 10 dB better than the signal you want to record, it degrades the signal by such a smidgen of a decibel that another 30 dB of signal-to-noise ratio will not get you a detectable change in the signal-to-noise of the program. It's a comparable sort of situation. There's no point in it.

CLARK: I think these arguments come from people who are trying to defend a belief that they have an emotional stake in, and they search around for anything that looks like technical support and just advance it. I don't think there's any logic behind it.

EARGLE: Dave, if you look at the amplifier business for the last 20 or 30 years, everybody has been advertising that they could pass a square wave with a fundamental frequency of 20 kHz, and they were damned and determined to show you that the thing retained a square wave shape all the way up, without any overshoot or with minimal rounding or something like that.

And they were the first ones who said that it's got to be this good in order to work in the passband of the ear.

LIPSHITZ: Which was fairly easy to do with amplifiers; you just make their band widths up in the hundreds of kHz range.

However, with digital systems to get that bandwidth is a little bit excessive!

EARGLE: And I don't know of anybody today who band-limits an amplifier to 20 kHz or anything near it.

LIPSHITZ: No, and that's why digital systems have come in for so much criticism because the transient signals, the square waves, don't look the way people are used to seeing them when they go through.

EARGLE: They look like sine waves. They are sine waves.

LIPSHITZ: But, you know, an experiment which we frequently do with students as a demonstration-I think that's on that ASA disc. Take a square wave of a fundamental frequency above 7 kHz so that-we believe that the true square wave is only odd harmonics-the third harmonic will be above 20 kHz; you can't hear it or any of the higher ones. So there should be no difference between that square wave and a sine wave of the same fundamental amplitude.

Now the fundamental of a square wave is not the same in peak-to-peak amplitude as the square wave is; it's bigger. I think the ratio is 4/m. That's 4 over 3-point-some thing, so it's a bit more than 1.

EDITOR: The amplitude 4 representing the sine wave and the amplitude 3.14 rep resenting the square wave?

LIPSHITZ: Yes. So, if the square wave was 1 volt peak to peak, the sine wave would have to be 4/m volts peak to peak. A bit more than one, if I remember the number correctly. I may be wrong; I think it's that.

And you've got to watch most function generators-when you switch square to sine, it keeps the peak amplitude the same, not the fundamental. So you will hear a difference solely on the basis of the level change. And if you do this experiment properly, people don't believe it. They look at it on the oscilloscope: "I can't hear the bloody difference!" There's nothing that will convince people more rapidly than something like that...

CARVER: But they'll go away thinking that perhaps you've tricked them.

LIPSHITZ: ...but they don't do that experiment! And when you ask about these rise times and so on-put the square wave through. Use a very good system for your transduction. If you could make one with a bandwidth of 100 kHz, you can have a rise time down in the microsecond range. So that means you're letting the first umpteen harmonics of the square wave into the ear

-you still won't hear the difference, un less you make the level so gross that you actually are creating significant intermodulation nonlinearities in the ear.

CARVER: And that's the only way you can hear 20 kHz. Because for somebody my age to hear 20 kHz requires a sound pres sure power about three trillion times greater than to hear 1 kHz, and at 15.750 kHz, the television oscillator, it requires about a mil lion times more sound power to hear than at 1 kHz.

LIPSHITZ: There is a benefit in not hearing the 15.750, mind you, isn't there? It helps not to hear that.

CARVER: Well, I can still hear that.

LIPSHITZ: Oh, you can? I can't. I can't hear that, either.

CARVER: Even a perfect ear-it's just marginal. You're only about 10 dB above the threshold.

EARGLE: You remember the London convention--oh, about seven or eight years ago--when KEF put on that demonstration of the audibility of various things?

LIPSHITZ: Clipping and that sort of thing?

EARGLE: Yes. And one test was the audibility of a noise signal extending beyond 20 kHz. It was a double-blind test.

LIPSHITZ: Oh yes. Yes.

EARGLE: And Laurie Fincham had come up with some sort of a ribbon tweeter, a

Japanese ribbon tweeter, that he mounted on these KEF speakers, so that the speakers actually went up that high-because his own speakers wouldn't do it. Anyhow, only one person out of the entire population statistically beat the odds, and it was an 18 year old guy from Denmark. He was the only one.

LIPSHITZ: Where was the cutoff of the filter for comparing?

EARGLE: Something like 23 or 24 kHz.

LIPSHITZ: Versus?

EARGLE: Twenty.

LIPSHITZ: Versus 20 kHz?

EARGLE: Yes. And he is the only one who could really hear it. Now, when you go from 20 down to 15, lots of people heard it.

LIPSHITZ: Oh yes, absolutely.

EARGLE: The idea being-we all have to ask ourselves this question-if one person out of a population of maybe, oh, let's say 250 members of the area who might have volunteered for these tests, if one person heard bandwidth in excess of 20 kHz, does that make it worth pursuing, intellectually, as engineers, or commercially? Probably no.

EDITOR: Don't say that too quickly.

EARGLE: Okay, I won't. But I think it's a point worth elaborating for the sake of our discussion here. What are our obligations? How do you feel about it?

CARVER: You're right. If it were possible to hear 22 or 23 kHz...

EARGLE: Yes, if those people could hear it, should we make records for them?

CARVER: ...if it were possible, and if they did hear it, I mean the SPL's must be just incredible at those frequencies, and it can't do anything but be hurting something, I would think.

EARGLE: No, it isn't that...

CARVER: I mean, you're going to be around at least 100 dB if you're going to hear 22 kHz...

LIPSHITZ: No, not necessarily.

EARGLE: No, at that age not at all.

LIPSHITZ: It depends. Some people do go beyond 20 kHz without enormous troubles.

EDITOR: Wouldn't you say that a perfectionist sound system should be designed for the most perfect ear among us? LIPSHITZ: How do we find that person? We first have to find that person and mea sure him or her before we can design the sound system then, Peter. So we have a bit of a problem. How many-3 billion people in the world now?

EDITOR: I could live with the most perfect one found so far.

LIPSHITZ: We better check them all quickly before we go any further.

EDITOR: "Found so far" I think is good enough.

CARVER: I've checked a lot of hearing and I've only found one person who can hear even 21 kHz, and again it was a very young person and everything was perfect.

And it required a tremendous sound pres sure before he would even hear it. I think I was overloading something.

LIPSHITZ: And in music, there is no musical content up there. Even if at a point you detected it...

EARGLE: Well, in normal music... I don't know. I don't know. When you talk about synthesizers...

LIPSHITZ: Well, if you know a composer who is composing for that one person...

EARGLE: Look, Wendy Carlos can make an arrangement of "Switched-On Bach" with an oscillator going all the way up to 25 kHz. Should it be heard or should it not be heard?

LIPSHITZ: Well, is she doing that because there is somebody who can hear up there and she wants him or her to hear it?

EARGLE: Well, let's assume so.

LIPSHITZ: Or just because she didn't have a suitable filter to put at the output of the oscillator? EARGLE: Let's assume that she is saying, if it can be heard, I'd like it to be heard. It's there to be heard.

LIPSHITZ: All right, I won't answer that.

I'll answer it with a question.

EARGLE: Okay. That's fair.

LIPSHITZ: Suppose Wendy Carlos was not being recorded and released on records.

Would you feel there's an obligation to record and release her work so that the one person who might be interested in hearing it can hear it?

EARGLE: Well, on the other hand, let me ask you a question. Let's say this is only on one piece and it's recorded as Band 1 on Side 1 of an LP...

LIPSHITZ: And that's not intended in any

"...has it been an arbitrary decision...that 20 or 21 kHz ought to be the upper limit-what we're going to be recording from now until the end of time?" way as a reflection on Wendy Carlos. This is just composer X.

EARGLE: Okay. Let's say this tone is re corded on Band 1 of one side of this LP, and you know it can be handled by that diameter. You can record to 35 kHz on the outer bands of an LP.

LIPSHITZ: Should you do so?

EARGLE: Well, you can play it back. It's there.

LIPSHITZ: Suppose you could record the RFI that your microphone line is picking up, without demodulating it, so that when you played your record back you could also get the FM broadcast that was going on in the background...

EARGLE: That would be a very wide range system...

LIPSHITZ: ...as a choice, instead of being forced to listen to the demodulated version.

(Everybody finds this hilarious.) For example, take Glenn Gould's Goldberg Variations, on CBS...

EARGLE: You hear a lot of musical externals going on.

LIPSHITZ: Yes, you hear plenty of Glenn Gould. But please tell me what the orchestra is that's playing in the background.

CLARK: Is there one?

LIPSHITZ: Oh yes.

EARGLE: Really?

LIPSHITZ: Yes. And I've had friends spend quite a few minutes trying to identify it. Because I wanted to write to CBS saying, "Why is Tchaikovsky's 5th playing in the background?" But we could not identify what it was.

CLARK: "And did you pay the royalties?"

EARGLE: Is that the early recording or the new one?

LIPSHITZ: No, no, no. The new one. The one recorded at CBS studios in New York City.

MCGRATH: The last Goldberg?

LIPSHITZ: If you want to try it, every body's got it. Yes, that one. Go to around 24 minutes.

EDITOR: I own that CD.

LIPSHITZ: Everybody owns it, man. CBS must be making a very good return on that.

EARGLE: Incredible.

EDITOR: I never noticed what you did.

LIPSHITZ: Listen around 24 minutes in the thing-that happens to be a good place to listen because the music's very quiet there

-and you will hear this orchestral thing going on in the background. Now it's throughout the disc; it comes and goes. It's almost certainly RFI; you will hear that be tween variations it fades down and comes back up, so you know they digitally edited the thing. Fade it down, bring it back up...

McGRATH: If it had been done in analog, I would have thought maybe they didn't quite degauss the tape; I've heard that on other things.

LIPSHITZ: It's on the master tape; almost certainly it must be coming in on their mike lines; it's RFI...

EDITOR: The next thing would have been "Breaker two! Breaker two!"

McGRATH: "Good buddy Glenn, how ya doin'? Come back, good buddy Glenn." (Laughter.)

EDITOR: That has happened to me, on my system, in the middle of the music.

LIPSHITZ: ...but if those mike preamps and the digital system had been processed linearly...that I can hear it...demodulate the radio station WQXR or whatever it was, then... [everybody talking at once, making this undecipherable- Ed.]

EARGLE: Okay. I see the point you're aiming at here. And it isn't quite what I'm talking about.

LIPSHITZ: I know, I know. (Laughs.) EARGLE: The thing is that we have cartridges and LP's and tape machines, at 30 ips, that will handle well beyond 20 kHz.

It's been shown that some people can hear some distance beyond 20 kHz. The question then becomes, basically, has it been an arbitrary decision, or why have we decided, that 20 or 21 kHz ought to be the upper limit-what we're going to be recording from now until the end of time? CARVER: Wait a minute. Wait, wait. It's been shown that some people can hear be yond 20 kHz but at such horrendous SPL's that it will never show up at that level. And if there's a 21 or 22 kHz musical bit on our recording, even that person is not going to hear it; it will be below his threshold at 22 kHz; he just won't hear it.

LIPSHITZ: But I appreciate John's point.

think my answer to you would be this. If there were no good engineering reasons for

wanting to restrict the bandwidth to some thing round 20 kHz, I'm sure we would have it wider.

CLARK: Just like amplifiers.

LIPSHITZ: Not because there are some people but just to give us a bit of extra lee way. Not because I happen to know one individual who could hear it. But the penalty, the engineering penalties...

EARGLE: It's an economic penalty.

LIPSHITZ: ...for increasing the bandwidth from 20 to 25 kHz in digital are extremely high. And given the choice that that might mean giving you 25% less playing time on your CD, for that one individual, I doubt that many people would make that decision.

EARGLE: Okay. I think you're probably absolutely right and I would say that it be comes a matter of the cost of real estate in the medium. And I would expect to see this particular point made in the transcript of this meeting we're having today. [Come on, John, every word is being transcribed:

that's why the damn thing is so long.- ed.]

LIPSHITZ: It's an engineering trade-off.

EARGLE: It is a trade-off. We would not choose to do it for any other reason.

McGRATH: But is that the reason it was chosen in the first place? I mean, why did Philips settle on the bandwidth that they did?

LIPSHITZ: Oh, there's a very good reason for that.

CLARK: That's an even number, and twice that is 44.1, which they already have...

LIPSHITZ: No. No. The reason is the following. The only feasible way of recording digital was on video recorders. If they hadn't established a format for recording on video recorders, there would have been no digital material available at the time of the CD release.

EARGLE: Soundstream was generating all the 50 kHz material, and there are a lot tapes around which have been transcoded over to 44.1.

LIPSHITZ: Right. But the question was, why was the CD 44.1. My answer is that the only practical standard that they could come up with-we're talking now...going on to a decade ago-was based on the video format, and that means you had to have an integral number of samples per horizontal line, and that led to 44.056 kHz in NTSC 59.94 fields-per-second countries-it would be 44.1 for 60-but anyhow, that's what led to that number. So it's just de facto;

if there had been other ways of storing it, I'm sure a much more attractive-looking number than 44,100 would have been chosen.

EARGLE: Yes, and you know, when you look at the entire standard, I've never been put off by the 16-bit limitation. That's never bothered me at all. But I have sort of wondered from time to time about the sampling rate because in our time, in this business, we've seen it: begin at 50, with Soundstream; we've seen it go to 48; and then we've seen it drop precipitously down to 44.1. One can only look at that and say, well, you better watch those bastards or they're liable to lower it again. It isn't going to happen, but for a while...

LIPSHITZ: There's a theorem that will pre vent them from lowering it anymore.

EARGLE: That's right.

EDITOR: It's like the fear of rising taxes- you have a fear of lowered sampling rates.

(Laughter) LIPSHITZ: But it's interesting... The funny thing is, John, you're happy to accept 16 bits, but you're reluctant to have too low a sampling rate, too close to the Nyquist limit.

EARGLE: Yeah. Yeah.

LIPSHITZ: ...yet, of those two trade-offs or engineering decisions, the former is the questionable one, and the latter is the iron clad one, in principle.

CLARK: Yes, that's right. That's right.

LIPSHITZ: If you can't hear above 20 kHz, you don't need a sampling rate more than two times that, plus whatever guardband you need for reasonable filter design.

EARGLE: Of course.

LIPSHITZ: Whereas the decision to use any finite number of bits, in principle, is a degradation, a loss. We can change what would be a distortion into a noise, by dithering or doing other things, but that's what fixes your signal-to-noise ratio; it is a trade off. The noise can be heard. So you've in troduced a noise, which in an infinite-bit system would not have been there.

EARGLE: Well, it's a noise which you can argue about, in terms of the dynamic range of music. The thing is that if you measure "My point is that, had consciences not prevailed here, we could have ended up with a medium that was 15-kHz limited and 14-bit limited." in any concert hall the maximum level of an orchestra and then look at the noise level into which it sinks when the music stops, you're nowhere near the inherent dynamic range limitation of a 16-bit system.

LIPSHITZ: True, but you're nowhere close to what the ear is capable of.

EARGLE: That's right.

CARVER: So when we combine what you said with what you said, we find that the existing digital system very comfortably fits totally inside the envelope of the human hearing mechanism, without being troubled at all.

EARGLE: Yes, it does.

McGRATH: No, no, no-the music replication mechanism, not the human ear.

LIPSHITZ: The music replication, yes...

McGRATH: The ear hears much lower than the 16 bit...

CARVER: Oh, yes, yes.

McGRATH: ...but the real world does not demand more than 16 bits.

LIPSHITZ: ...what I'm saying is, the 16 bit decision has introduced a noise floor which in realistic music, as John is saying, is not a limitation. He can't find a hall that's quieter to record in. However, if I wanted to have a recording system that would never add its own noise to any re cording I made, I need a dynamic range of at least 120 dB-because that's the approximate dynamic range of the ear. And that means that I need more than 16 bits. Where does that take us up to...? 18, 20... We need about 20 bits to do that. And then you get very dicey, if you need more than that.

Because then you're designing for creatures we haven't found yet, who maybe will visit us from outer space one day-and you wouldn't want to limit their dynamic range... (Laughter)

CLARK: Well, we should have some music for them...

EARGLE: I suggest then that we further state that the decision to use 16 bits is in it self an unfortunate compromise and say that we around this table all wish that these limits could be extended.

LIPSHITZ: No, I'm not sure I wish that one.

CLARK: I don't wish to pay for it.

EARGLE: Well, okay. It sound to me as though everybody here is in agreement that Philips or Sony-and I don't trust those bastards any further than I could throw them (spoken with a facetious inflection)- made the right compromise for mankind forevermore. Now, I'm sure you don't think that.

LIPSHITZ: Now wait. We've already discussed the sampling rate, and I think we said, all else being equal the engineers would probably have chosen a higher one if they'd been able to (inaudible word) for it.

CARVER: For comfort reasons...

EARGLE: And they probably would have chosen more than 16 bits if they had the leave to...

CARVER: ...for just sort of comfort reasons, but not for any real demonstrable reasons.

LIPSHITZ: Well, look. The original Philips conception was 14 bits. It was as a result of Sony's prodding that it was raised to 16- because it didn't require unacceptable loss of playback time on the medium, thanks to clever channel encoding and so on.

EARGLE: I must say something here. I wanted to talk about unfortunate compromises.

LIPSHITZ: Yes, let him modify his statement, then I'll comment on it.

EARGLE: I won't modify it; I'll elaborate.

I'm going to say that if we had had Philips succeeding with their 14-bit suggestion here, and somebody else were saying that, well, FM, which our best medium in many ways, is limited to 15 kHz, so why not go to 32 kHz sampling-and where do you think that unfortunate number came from? LIPSHITZ: No, that I call unfortunate.

EARGLE: Okay.

LIPSHITZ: 14-bit I would call unfortunate because it's not significantly better than analog tape. 16-bit I don't call unfortunate.

EARGLE: Okay. My point is that, had con sciences not prevailed here, we could have ended up with a medium that was 15-kHz limited and 14-bit limited.

LIPSHITZ: Yes.

CARVER: That's true. We came close to that.

LIPSHITZ: We came close.

CARVER: We came close.

LIPHITZ: But it's like the slew-rate question--isn't it?--with amplifiers. How close can you come to the slew rate before it's unacceptable...

CARVER: It's like falling off a cliff.

LIPSHITZ: ...how much margin do you need?

CARVER: As long as you don't fall off that cliff, you're perfectly okay. You can come as close as you please.

LIPSHITZ: There are people who believe you want a slew-rate margin of a factor of 10 so that you never come close to the limit. There are other people who believe that you can come up to 0.9 of the slew limit and not be able to tell that you're there.

EARGLE: And there are people who will tell you that a bridge ought to be built with 10 times safety factor in every area-and it's a prohibitively expensive bridge but it will never crash.

LIPSHITZ: But they never get the contract because their price is too high.

EARGLE: That's right.

LIPSHITZ: But anyhow, on this number of-bits question, I would say the following.

A 16-bit system gives a 98 dB signal-to noise ratio in principle, undithered.

EDITOR: 98.1, right?

EARGLE: An even wider dynamic range.

LIPSHITZ: The dynamic range is wider be cause you can hear things below the noise.

But that's true of any system. In practice, in a properly dithered system, you're talking more than 90 dB. Now, for most people, it is really quite eye-opening to play some thing very loudly, perhaps significantly be yond the level they would ever want to play music at, something corresponding to full modulation of the medium, and then play them a -90 dB tone-and they'll say, you haven't played anything; they'll say there's nothing on. That's what they'll say. Then you say, go to the loudspeaker, put your ear there. Oh yes, ah! You know, 90 decibels is really...

EARGLE: It's a lot.

LIPSHITZ: People don't quite appreciate it. Now, yes, the noise floor in a quiet room like my listening room is just at the level of the subliminal. You can perceive the noise floor of the digital system; it's not below my threshold. It would be nice if it were be low; a few dB more might be nice. But, goodness, for a medium designed for hundreds of millions of people out there at an affordable price, it's not bad. It's pretty damn good. Now, the professional I think needs more than 16 bits because he needs to be able to do some manipulations and mixing and headroom and other things, and be able to get 16-bit signal-to-noise in the end.

EARGLE: Are you going to go on record with that statement?

LIPSHITZ: Yes.

EARGLE: Good.

CARVER: Now I understand what your objection was. And you just brought it out, Stanley. Because you need that headroom.

You'd love to know that you're 12 dB away from clipping, or something like that, when you're operating your machines or editing your recordings. And the person who makes the final copy that we as consumers will listen to, he...

McGRATH: Will get the full 16.

CARVER: I suppose, I imagine that they somehow look through the thing, find the loudest passage, and park it right up at the top-and he has his 98 dB.

LIPSHITZ: Not just that, but multitrack digital.

CARVER: But when you're making the re cording you don't have that luxury; you need the extra 10, 12, 15, maybe 20 dB, to make sure you never run out of headroom.

LIPSHITZ: You want some at the top, and it's also good to have some at the bottom, so when you're mixing twelve channels together, you don't get twelve 16-bit noise floors added together but twelve 18- or 20 bit noise floors.

CARVER: So that's where your 20-bit would be, yes.

CLARK: Essentially, there should be a difference between pro and consumer machines then.

CARVER: Yes.

EDITOR: What is the widest analog dynamic range that you're aware of? Or the lowest analog noise floor-whichever way you want to put it. Isn't it around 80? EARGLE: No. More than that. I'm allowing the use of very artful noise reduction. I would say that normal 15-ips tape, with Dolby SR, is going to crack 100.

EDITOR: So that's 100 versus 98.1.

CLARK: Well, then there's dbx, too, which 18h EARGLE: There's dbx, which is much more audible, more of the time.

CLARK: But still, 120 dB.

EARGLE: I've never heard a peep out of SR.

(This is where we must pause. To be continued in these smaller installments.)

--------

[adapted from TAC, Issue No. 15]

---------

Also see:

Box 978: Letters to the Editor

Various audio and high-fidelity magazines