Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: "Putting the Science Back into Loudspeakers" (Read 9439 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

"Putting the Science Back into Loudspeakers"

http://www.celticaudio.co.uk/articles/science.pdf

I have read the article, and this guy has a few interesting (although well-known) points about loudspeaker directivity and the connection between doing a recording without knowing the playback system vs constructing/installing a hifi system without knowing the recording technique.

The way that the author use "information theory" seems strange though. A few other things also seems less thought-through:
Quote
We have reached the unsupportable position where loudspeakers are frequently not good enough
to assess the quality of other audio components. This leads to flawed experiments whose
outcomes are unreliable. The most serious effect has been the widespread adoption of lossy audio
coding algorithms. This must be at least partly due to the fact that their flaws cannot be heard on
the legacy loudspeaker.

And this has to be because of bad speakers? What if the speakers are good enough, but the listeners cannot hear the difference, or simply doesnt care?

Quote
Traditional loudspeaker measurements are disreputable. It is widely known that several speakers
having the same measurements can sound quite different.

Is it possible to measure a pair of speaker in all ways that concievably matters to sound reproduction for humans, find that they are "identical" for all parameters well within the thresholds that are considered to be JND, and still find that a well-organised blind-test reveals audible differences?

Quote
Another area in which loudspeakers are disreputable is in the neglect of the time domain. The
traditional view is that all that matters is to be able to reproduce continuous sine waves over the
range of human hearing.
A very small amount of research and thought will reveal that this is a misguided view. Frequency
response is important, but not so important that the attainment of an ideal response should be to
the detriment of realism. One tires of hearing that "phase doesn't matter" in audio or "the ear is
phase deaf". These are outmoded views which were reached long ago in flawed experiments and
which are at variance with the results of recent psychoacoustic research.

For Linear time-invariant systems, the fase/frequency respons actually has the same information as the time-response.

Engineers commonly "disregard" the phase-response, but this is a conscious choice of the designer based on what he believes to be important.
Quote
Lack of attention to the time domain in
crossover networks leads to loudspeakers which reproduce a single input step as a series of
steps, one for each drive unit at different times. The use of resonance in reflex cabinets masks the
relaxation time in the audio signal.

But proving this may not be easy. Perhaps the designer should spend the limited funds on other things than compensating for a bass-design less efficient than bass-reflex, or time-aligning elements through complex enclosure design? I dont know.
Quote
Considering information theory, a steady state sine wave carries no information because it has no
bandwidth and any one cycle is predictable from the one before. Only transients have bandwidth
and contain information because they are unpredictable. It follows that a speaker which is
optimized to reproduce steady state sine waves does not necessarily have adequate information
capacity.

For a purely linear system (a good model for good loudspeakers at sufficiently low levels), a loudspeaker that can reproduce any sine at any frequency within a given bandwidth at the right phase and level will have quite a lot of information capacity. Shall we consult Shannon?
Quote
The same is true for compressors or bit-rate reducers. It follows that codecs can only meaningfully
be assessed on speakers of adequate information capacity. It also follows that the definition of a
high quality speaker is one which readily reveals compression artifacts. The only audio quality
criteria we have for sound reproduction is that performance actually meets psychoacoustic
requirements.

I believe that audio codecs are often evaluated using headphones. Since they have an impulse-response (i.e. time-response) closer to "ideal" than any loudspeaker in a room could realistically aim for (at least as long as we leave digital room/speaker-correction out of the picture)

I dont think that you can define good speakers from how good they reveal compressed material, be it dynamically compressed or digital bitrate-compressed.
Quote
Lossy compression does not preserve the original waveform and seeks to be blameless by placing
the noises where they will be masked. Naturally one would want to carry out listening tests to see
if this goal had been achieved. If blameless loudspeakers are used, the test is valid. However, the
legacy loudspeaker is not blameless and does not preserve the waveform either

This is correct, but if you do blindtests at the homes of audiophiles with expensive (=good?) systems and a high interest in audio (commonly also a desire to discredit lossy audio), and these cannot confirm audible differences, then it seems natural to assume that:

"This codec at this bitrate does not seem to reduce the persepted audio quality for any of a wide range of listeners and playback systems, including high-end ones"

Of course, new advances in loudspeaker design and perception could concievably lead to lossy differences being audible in 10 years, but does that in any way affect our choice of "transport format" used for mobile and casual sound reproduction today?
Quote
MPEG layer 2 and Dolby AC-2 coders even though their internal workings is quite different. In
retrospect this is less surprising because both are probably based on the same psychoacoustic
masking model. MPEG-3 fared even worse because the bit rate is lower.

What is MPEG-3, and what general audio codec has a "lower bit rate" per definition?
Quote
The effects are not subtle and do not require "golden ears". We have successfully demonstrated
these effects to an audience of about 60 in a conference room on more than one occasion;

But still no blind-tests?

-k

"Putting the Science Back into Loudspeakers"

Reply #1 –
The problem is always the same: Over- and underestimation of one's own consciousness. People dont understand themselves. This is overally will not go away, because it is a civilization-wide problem with peoples mentality and lack of self-understanding. Fixing this isn't limited to simply distributing specific information - generally changing this, would require a civilization-wide shift in self-understanding. That doesn't sound like something which is gonna happen soon.

Therefore, i highly doubt that "information" is an efficient strategy against that in the short- and midterm. Circumvention- and neutralization-tactics may be more efficient (just like the loudness-race is best fought with mass-distribution of replaygain).
I am arrogant and I can afford it because I deliver.

"Putting the Science Back into Loudspeakers"

Reply #2 –
As has been pointed out here on numerous occasions, often artifacts of lossy encoding will be most apparent on low-quality reproduction systems or to people who have specific hearing problems. The reason should be obvious - the psychoacoustics assume normal hearing through a system that does not distort or color the sound in an unexpected way. Also, tuning of the psy model is very often done based on tesing using high quality headphones, so are most likely to be transparent when listened to in this way.

"Putting the Science Back into Loudspeakers"

Reply #3 –
Quote from: pdq on
As has been pointed out here on numerous occasions, often artifacts of lossy encoding will be most apparent on low-quality reproduction systems or to people who have specific hearing problems. The reason should be obvious - the psychoacoustics assume normal hearing through a system that does not distort or color the sound in an unexpected way. Also, tuning of the psy model is very often done based on tesing using high quality headphones, so are most likely to be transparent when listened to in this way.

But since the referred information was coined as a serious article, I thought it was important to comment it.

-k

"Putting the Science Back into Loudspeakers"

Reply #4 –
knutinh,

The parts you quoted about loudspeaker design, measurement and performance all seem accurate, and I think you're misunderstanding them, or ignoring important words, in judging them so harshly.

The parts you quoted about lossy coding lack detail.

Cheers,
David.

"Putting the Science Back into Loudspeakers"

Reply #5 –
The guy's name is Watkinson, thread title should be fixed.

His disinclination to mention blind testing even once, while reporting results of what appear to be from dubious 'sighted' comparisons, especially as regards evaluating lossy codecs, raises all sorts of red flags to me, despite his credentials.

"Putting the Science Back into Loudspeakers"

Reply #7 –
Quote from: 2Bdecided on
knutinh,

The parts you quoted about loudspeaker design, measurement and performance all seem accurate, and I think you're misunderstanding them, or ignoring important words, in judging them so harshly.

The parts you quoted about lossy coding lack detail.

Cheers,
David.


Quote from: knutinh on
Quote

We have reached the unsupportable position where loudspeakers are frequently not good enough
to assess the quality of other audio components. This leads to flawed experiments whose
outcomes are unreliable. The most serious effect has been the widespread adoption of lossy audio
coding algorithms. This must be at least partly due to the fact that their flaws cannot be heard on
the legacy loudspeaker.


I think this is a conclusion not supported by the argumentation in the text.

The argumentation does not support that the widespread adoptation of lossy audio coding must be (at least partly) due to flaws present in loudspeakers. In fact, it could be _only_ due to the "limited capacity" of our hearing, or it could simply be due to teenagers not caring about sound quality compared to mobility and price.

Quote
Quote

MPEG layer 2 and Dolby AC-2 coders even though their internal workings is quite different. In
retrospect this is less surprising because both are probably based on the same psychoacoustic
masking model. MPEG-3 fared even worse because the bit rate is lower.


I think if you are going to write a white-paper (?) partially about listening tests done on lossy audio-codecs, it would be wise to ensure that the codec names you are referring to actually exists. Referring to non-existant codecs makes the reader sceptical about other content as well.

http://en.wikipedia.org/wiki/MPEG-3

I would also expect a recent paper to use a codec newer than the Dolby AC-2 codec (1989?)

http://www.computerhope.com/jargon/a/ac3.htm

-k

"Putting the Science Back into Loudspeakers"

Reply #8 –
I didn't seek to defend the woolly parts on lossy coding. The author is clearly straying outside his area of expertise.

I was defending quotes 2-5 in your original post. The parts you quoted are correct - your comments are not. I'm not being rude - I'm just trying to avoid taking the time to unpick everything you said - I'm sure Google could help instead.


Anyway, from the quotes, the basic premise he makes appears to be an interesting one.

Lossy coding is based on the fact that the human ear itself is lossy. However, all the psychoacoustic measurements ever made on human ears must, by necessity, use real transducers: headphones or speakers. As we all know, these transducers are often the least perfect part of the audio reproduction chain. The levels of distortion,  colouration (non flat frequency response), and time domain distortion found in the very best loudspeakers can be far worse than you would ever find in a $5 amplifier!

So in reality, with every psychoacoustic test, you are measuring the limits of a system - and that system is the amplifier, speaker/headphone, human ear and human brain.

His argument is simple: the speaker is known to be poor. The human ear is believed to be even poorer, and so it is assumed that the data is good enough. Obviously it will be good enough if we continue to use the same kind of speakers.

However, it may be some of the data is due to the transducer, rather than the ear. If a different (better) transducer was invented, all the previous data would tell the psychoacoustic model that something was inaudible, or masked that was in fact only hidden by distortions and deficiencies in the old transducer.


There are arguments against this, but the most often quoted one is bogus: yes, Dibrom did tune lame --alt-preset standard using a laptop and crappy earphones, but the kind of pre-echo artefacts which plague(d) lame mp3 are most audible via headphones (good or bad) and least audible via speakers. However, these aren't the only kind of artefacts: high frequency loss is (quite obviously) less audible via transducers that attenuate high frequencies. So it is quite conceivable that certain types of noise and distortion would be less audible through transducers that generated similar noise and distortion.


The argument can be debunked if it can be shown that good loudspeakers already produce distortion that is (in every sense) an order of magnitude (or more) lower than that introduced by high quality psychoacoustic coding.

I suspect this is true, but it's not a trivial thing to prove.

Usefully, it is sometimes proven for individual psychoacoustic tests, where the output from the transducer is captured using a high quality measurement microphone, and analysed to prove that there are no spurious components in the signal which could interfere with the results.

This wasn't always the case: in the earlier "absolute threshold of hearing" and "equal loudness" experiments, the curves were simply wrong at frequency extremes because subjects were hearing distortion (harmonics or subharmonics) as well as the intended frequency, thus giving them an extra clue that the sound was present, or increasing the perceived loudness.

To prove that something similar is not happening with modern psychoacoustic coding, you would need to measure the distortion introduced by a good loudspeaker on real music, and demonstrate that this was significantly lower on every time and frequency scale, than the changes introduced by a good (transparent) psychoacoustic codec. Measuring distortion with a real music signal is very difficult, but possibly the analysis exists to account for the linear frequency response, and the effect of the room (an anechoic chamber will help), leaving just the distortion. If not, artificial test signals must be used (something like the belcher distortion test), but that makes is harder to draw a confident conclusion that speakers produce lower distortion than psychoacoustic coding for real music signals.


The alternative is that we wait for these new transducers to arrive, and then listen to them to see if psychoacoustic coding falls down. I suspect it will work about as well as it does now, and we all have bigger things to worry about.

Cheers,
David.
 

"Putting the Science Back into Loudspeakers"

Reply #9 –
Quote from: 2Bdecided on
I didn't seek to defend the woolly parts on lossy coding. The author is clearly straying outside his area of expertise.

Thank you.
Quote
I was defending quotes 2-5 in your original post. The parts you quoted are correct - your comments are not. I'm not being rude - I'm just trying to avoid taking the time to unpick everything you said - I'm sure Google could help instead.

I know that you have a lot of knowledge. I still cant see how my comments are wrong, or any way to let google figure out. The way I read this paper, the author is misinformed about the connection between frequency/phase-response and time-behaviour for LTI systems. He has some controversial, unproven statements regarding measurements correlation to subjective response, and he seems to have forgotten that headphones solve many of his philosophical dilemmas.

I think that involving "fancy words" picked up from information-theory should include a more thorough review so as not to look like "word-dropping" in order to look knowledgeable.
Quote
Lossy coding is based on the fact that the human ear itself is lossy. However, all the psychoacoustic measurements ever made on human ears must, by necessity, use real transducers: headphones or speakers. As we all know, these transducers are often the least perfect part of the audio reproduction chain. The levels of distortion,  colouration (non flat frequency response), and time domain distortion found in the very best loudspeakers can be far worse than you would ever find in a $5 amplifier!

If you want an example of the effect you are describing, I think that there are some references to the 1930s, where it was claimed that frequency response > 6kHz was detrimental to sound. The reason evidently was that low-bandwidth loudspeakers suppressed non-linear distortion found in much of the equipment.

In practical terms, if it cant be heard today, then it would be an arguement for using lossy codecs right now anyways. It could, however be an arguement for designing better speakers (and thereby needing better signal sources), but that would be a claim that should be argued by the author. It seems that the author is constructing loudspeakers for sale, and probably tries to follow his own principles in loudspeaker design. A good proof therefore would be a DBT of his speakers combined with high-quality lossy codecs, showing beyond doubt that those loudspeaker design philosphys reveal codec artifacts invisible in other systems. No such DBT is mentioned.

-k

"Putting the Science Back into Loudspeakers"

Reply #10 –
OK, I'll bite...

[quote name='knutinh' post='538533' date='Jan 1 2008, 14:28']
Quote
Traditional loudspeaker measurements are disreputable. It is widely known that several speakers
having the same measurements can sound quite different.
Is it possible to measure a pair of speaker in all ways that conceivably matters to sound reproduction for humans, find that they are "identical" for all parameters well within the thresholds that are considered to be JND, and still find that a well-organised blind-test reveals audible differences?[/quote]He didn't say that at all. He said "traditional loudspeaker measurements" by which, I assume, he means frequency response (usually third octave averaged), impedance (usually a single misleading number), and power (usually meaningless).

Even if he'd being more subtle, you can have identical on-axis measurements, but performance in real rooms being dramatically different due to different off-axis response. It is so complex that the best solution is usually to listen to the speaker you want in the room you want it in.


Quote
Quote
Another area in which loudspeakers are disreputable is in the neglect of the time domain. The
traditional view is that all that matters is to be able to reproduce continuous sine waves over the
range of human hearing.
A very small amount of research and thought will reveal that this is a misguided view. Frequency
response is important, but not so important that the attainment of an ideal response should be to
the detriment of realism. One tires of hearing that "phase doesn't matter" in audio or "the ear is
phase deaf". These are outmoded views which were reached long ago in flawed experiments and
which are at variance with the results of recent psychoacoustic research.
For Linear time-invariant systems, the fase/frequency respons actually has the same information as the time-response.
He didn't say the frequency/phase response - he said the frequency response - that means (typically) the amplitude/frequency response - nothing more. He clearly means to criticise this approach by criticising those that ignore the phase. Of course frequency-amplitude plus frequency-phase is perfectly mathematically related to the time domain response as you say - but given just two frequency-domain graphs (amplitude and phase), it's not easy to imagine the time domain (impulse or step) response from those graphs. The waterfall response plot is also related, but best seen directly, not guessed at from looking at the other graphs (though a reasonable guess can often be made).


Quote
Engineers commonly "disregard" the phase-response, but this is a conscious choice of the designer based on what he believes to be important.
Exactly - but in the 21st century, you'd think we could make loudspeakers which got both roughly correct.


Quote
Quote
Lack of attention to the time domain in
crossover networks leads to loudspeakers which reproduce a single input step as a series of
steps, one for each drive unit at different times. The use of resonance in reflex cabinets masks the
relaxation time in the audio signal.
But proving this may not be easy. Perhaps the designer should spend the limited funds on other things than compensating for a bass-design less efficient than bass-reflex, or time-aligning elements through complex enclosure design? I dont know.
I don't know if cost is an issue here - I'll have to check the website!


Quote
Quote
Considering information theory, a steady state sine wave carries no information because it has no
bandwidth and any one cycle is predictable from the one before. Only transients have bandwidth
and contain information because they are unpredictable. It follows that a speaker which is
optimized to reproduce steady state sine waves does not necessarily have adequate information
capacity.
For a purely linear system (a good model for good loudspeakers at sufficiently low levels), a loudspeaker that can reproduce any sine at any frequency within a given bandwidth at the right phase and level will have quite a lot of information capacity. Shall we consult Shannon?
But again you twist his words. You could make a speaker which can reproduce a sine wave well, and all frequencies individually at roughly equal volume (so producing a nice frequency-amplitude graph), but the high frequencies could be delayed by seven second relative to the low frequencies. Or a certain range of frequencies could ring like a bell, and decay slowly after the source stopped. If all you optimise is the frequency-amplitude response, you might have a very strange speaker.


Quote
Quote
The same is true for compressors or bit-rate reducers. It follows that codecs can only meaningfully
be assessed on speakers of adequate information capacity. It also follows that the definition of a
high quality speaker is one which readily reveals compression artifacts. The only audio quality
criteria we have for sound reproduction is that performance actually meets psychoacoustic
requirements.

He's heading off course here, but still...
Quote
I believe that audio codecs are often evaluated using headphones. Since they have an impulse-response (i.e. time-response) closer to "ideal" than any loudspeaker in a room could realistically aim for (at least as long as we leave digital room/speaker-correction out of the picture)
The ear is quite capable of hearing "through" the visually (on a waveform display) bizarre impulse response of a room to pick out near-invisible (on a same scale waveform display) pre-echoes from mp3 encoding (for example).

Quote
I dont think that you can define good speakers from how good they reveal compressed material, be it dynamically compressed or digital bitrate-compressed.
Agreed. Bad loudspeakers could hide this (having no speaker at all will hide it very well), but a speaker which shows it clearly might not be "good" - maybe it boosts everything over 16kHz by 50dB so even I can hear what mp3 does up there!

Quote
Quote
Lossy compression does not preserve the original waveform and seeks to be blameless by placing
the noises where they will be masked. Naturally one would want to carry out listening tests to see
if this goal had been achieved. If blameless loudspeakers are used, the test is valid. However, the
legacy loudspeaker is not blameless and does not preserve the waveform either
This is correct, but if you do blindtests at the homes of audiophiles with expensive (=good?) systems and a high interest in audio (commonly also a desire to discredit lossy audio), and these cannot confirm audible differences, then it seems natural to assume that:

"This codec at this bitrate does not seem to reduce the persepted audio quality for any of a wide range of listeners and playback systems, including high-end ones"

Of course, new advances in loudspeaker design and perception could conceivably lead to lossy differences being audible in 10 years, but does that in any way affect our choice of "transport format" used for mobile and casual sound reproduction today?
To turn that around: some people rip to lossless. but other rip to lossy at a quality which they believe to be transparent. Should they expect to re-rip when someone invents a new loudspeaker technology? If no, how can you be sure? If yes, then you agree with the writer you're quoting!

Quote
Quote
MPEG layer 2 and Dolby AC-2 coders even though their internal workings is quite different. In
retrospect this is less surprising because both are probably based on the same psychoacoustic
masking model. MPEG-3 fared even worse because the bit rate is lower.
What is MPEG-3, and what general audio codec has a "lower bit rate" per definition?
Quote
The effects are not subtle and do not require "golden ears". We have successfully demonstrated
these effects to an audience of about 60 in a conference room on more than one occasion;
But still no blind-tests?
He's lost it now, though I'm sure both of us are capable of producing mp3s that sound sufficiently terrible to be audibly poor, even over a train station PA system!

Cheers,
David.

EDIT: I didn't think there was much chance of getting all the quotes right, but I tried!

"Putting the Science Back into Loudspeakers"

Reply #11 –
Quote
We have reached the unsupportable position where loudspeakers are frequently not good enough
to assess the quality of other audio components.


But speakers have always been the weakest link.  For decades we've had the ability to design amplifiers with less than .1% THD and frequency response that's flat to within 1 dB 20Hz-20KHz.  Likewise, professional studio tape decks back in the 70's, and more modern digital recording technology since then could match those numbers.  But no commercial speakers operating at reasonable power levels in real-life listening environments can do so.  The sheer physics of efficiently moving a mass of air over a wide range of frequencies with paper/plastic electromagnetic contraptions in room-sized acoustic environments with a variety of reflective and absorptive surfaces makes it impossible.

"Putting the Science Back into Loudspeakers"

Reply #12 –
BTW, I am not trying to turn this into an arguement. I find that discussing things like this is interesting and contributes to a better understanding. If at any time you feel differently, feel free to ignore me :-)

This forum seems to react badly to long, deeply quoted messages, as I have two separate posts in my copy/paste-buffer that render flawless in preview-mode, but when I paste both sequentially, they are rendered as plain-text...
Quote from: 2Bdecided on
OK, I'll bite...

Quote from: knutinh on

Quote

Traditional loudspeaker measurements are disreputable. It is widely known that several speakers
having the same measurements can sound quite different.

Is it possible to measure a pair of speaker in all ways that conceivably matters to sound reproduction for humans, find that they are "identical" for all parameters well within the thresholds that are considered to be JND, and still find that a well-organised blind-test reveals audible differences?

He didn't say that at all. He said "traditional loudspeaker measurements" by which, I assume, he means frequency response (usually third octave averaged), impedance (usually a single misleading number), and power (usually meaningless).

Even if he'd being more subtle, you can have identical on-axis measurements, but performance in real rooms being dramatically different due to different off-axis response. It is so complex that the best solution is usually to listen to the speaker you want in the room you want it in.


I would say that even traditional loudspeaker measurements at least include 30 degrees off-axis response, and usually the power response.

Since he is talking in a "scientific context", i.e. in effect stating that it is impossible to know if mp3 is audibly degrading sound because even academia cannot characterise the sound of loudspeakers, I guess I expected "traditional measurements" to include a little more than the over-smoothed on-axis response shown in marketing material. That is not to say that characterising loudspeaker perseptual quality using measurements is a simple exercise.

Quote from: 2Bdecided on
Quote from: knutinh on

Quote

Another area in which loudspeakers are disreputable is in the neglect of the time domain. The
traditional view is that all that matters is to be able to reproduce continuous sine waves over the
range of human hearing.
A very small amount of research and thought will reveal that this is a misguided view. Frequency
response is important, but not so important that the attainment of an ideal response should be to
the detriment of realism. One tires of hearing that "phase doesn't matter" in audio or "the ear is
phase deaf". These are outmoded views which were reached long ago in flawed experiments and
which are at variance with the results of recent psychoacoustic research.

For Linear time-invariant systems, the fase/frequency respons actually has the same information as the time-response.

He didn't say the frequency/phase response - he said the frequency response - that means (typically) the amplitude/frequency response - nothing more. He clearly means to criticise this approach by criticising those that ignore the phase. Of course frequency-amplitude plus frequency-phase is perfectly mathematically related to the time domain response as you say - but given just two frequency-domain graphs (amplitude and phase), it's not easy to imagine the time domain (impulse or step) response from those graphs. The waterfall response plot is also related, but best seen directly, not guessed at from looking at the other graphs (though a reasonable guess can often be made).

I believe that even recent research fail to support the views of certain audio designers claiming that low-order phase-linearity is of great importance.

To reproduce continous sine-waves over the range of human hearing, I would say that both phase and amplitude has to be right.

This may seem like nit-picking, but given the enormous amount of people not understanding the relation between time and frequency, I think that any paper adressing the general consumer should be really clear on this topic.

Quote from: 2Bdecided on
Quote from: knutinh on

Engineers commonly "disregard" the phase-response, but this is a conscious choice of the designer based on what he believes to be important.

Exactly - but in the 21st century, you'd think we could make loudspeakers which got both roughly correct.

If the cost of getting it "right" is very low and/or it is significant, then yes.

If it is an unnecessary burden on the cost of the product that only has any meaning to those measuring loudspeakers, then I'd think it was wise to disregard it.

BTW, I won't claim to know what level of phase-distortion matters. I made a couple of phase-linear (in the system-sense, of course) xover-filters for my thesis that did not fare all that well in listening tests, even though the theory seemed sensible:-)

The november-issue of JAES included some listening tests to FIR and IIR-crossoverfilters of different phase-properties.

-k

"Putting the Science Back into Loudspeakers"

Reply #13 –
Quote from: knutinh on
I would say that even traditional loudspeaker measurements at least include 30 degrees off-axis response, and usually the power response.


Yes, I was wondering what Watkinson actually meant by 'traditional'?  Audio magazine bench results from 15-20 years ago? Because I would guess that the 'tradition' since then has come to include several more measurements.  Seems a straw man to me.

"Putting the Science Back into Loudspeakers"

Reply #14 –
I find the whole premise of reduced distortion as necessarily increasing artifact audibility as being fundamentally flawed. I think this has already been alluded to in this thread, but it needs to be drilled into people's heads. (I will grant that I have banged on this particular drum before.) Two counterexamples, off the top of my head:
  • Gross frequency response distortions, as David mentions. This isn't even merely theoretical: Anybody remember that c't MP3 DBT, using Sennheiser Orpheus headphones, where the only listener who got a positive result had major hearing damage? Nobody would claim that he could hear "better" than anybody else, merely that his particular hearing response managed to unmask <8khz artifacts.
  • Even-order harmonic distortion are believed to increase the audibility of absolute polarity differences, so that higher-distortion amplifiers (notably, tubes) would show greater sensitivity to that "artifact" than would a straight-wire, perfectly ideal amplifier. The same effect would presumably apply for transducers.
In the most general case, it's not at all proven that all reductions in distortion necessarily result in a reduction of ATH, or any other increase in sensitivity. If you have a "perfect" hearing model that accurately represents the playback chain as well as the listener's hearing, you have much more information available to make a more informed (and aggressive) psyacoustic compression. Given this hypothetically "perfect" codec, no DBT would ever succeed for that playback/listener combination - but it could easily fail for other listeners or other playback chains.

Long story short, we already know situations where distortions raise artifact sensitivity, and they are not particularly easy to identify a priori, so using artifact sensitivity as a metric for playback chain quality is dangerous.

"Putting the Science Back into Loudspeakers"

Reply #15 –
I agree overall with the posters who find this article to be scientifically incomplete. I will add one more element to the mix for fun.

Way, way back in college I was a TA for Dr. Amar Bose (I will pause here until the heckling dies down). Whatever one's opinion of his company may be, he was in fact a terrific teacher who exuded enthusiasm for his subject.

We used to get into discussions about audio fidelity, and he would laugh at some of the more esoteric products being sold - such as solder so fabulous that even changing one joint in your preamp would alter your listening experience. He appreciated that the human ear was very sensitive to certain things and amazingly forgiving about others, and saw his job as focusing upon those things that made what he considered to be positive perceptive changes.

Though we disagreed on many things, he convinced me of one right away. Simply stated, stereo recording is a sham with no hope of ever sounding like a convincing live event. Worrying about things like small changes in phase alignment (which are barely perceptible by the great majority) is nothing when compared with the elephant in the room: you have only 2 sources and your recordings are not made correctly. They can’t be, for now.

Why do we have 2 channel stereo as our current standard? Not because it is scientifically proven to work best - we have it because it is more fun than mono and was economically and technically plausible around 1959. Really, that is the answer.

Dr. Bose argued that the ear is very sensitive to phase in the context of spatial data – but this data is lost or mangled when only 2 sources are used. Furthermore, there is absolutely no guarantee that the recording is even made in such a way that the phase data is usable no matter how many channels are used. Close mic’ing, multiple mics, overdubs, instruments placed here and there relative to one another, electronic instruments – the list of recorded sounds that fail to communicate useful spatial data is endless.

Now take that data, reduce it even further to a 2 channel mix and play it back in your living space with real speakers, real walls and real furniture and real people. Reflections, time delays with random relationships to frequency due to the chaotic nature of the objects in the room – you get the picture. Unless you had a recording and playback system (viewed as a unified whole) that encompassed the entire space around the listener, you had little hope of reliably creating anything that sounded “real”. Perhaps once in a while you would be fooled, but not for more than a few seconds.

So, given the realities of the products that will be used with speakers (LPs, tape and CDs) he felt that the best bet was to create a system with a relatively uniform power response that would interact with the room to eliminate most “dead spots” and be fun to listen to. Phase response be damned, it’s a red herring. That’s the 901.

So, way back to the original thread – maybe it isn’t loudspeakers that are the weak link. It is the whole system from one end to another.

"Putting the Science Back into Loudspeakers"

Reply #16 –
Quote from: BradPDX on
Though we disagreed on many things, he convinced me of one right away. Simply stated, stereo recording is a sham with no hope of ever sounding like a convincing live event. Worrying about things like small changes in phase alignment (which are barely perceptible by the great majority) is nothing when compared with the elephant in the room: you have only 2 sources and your recordings are not made correctly. They can’t be, for now.

Why do we have 2 channel stereo as our current standard? Not because it is scientifically proven to work
So, given the realities of the products that will be used with speakers (LPs, tape and CDs) he felt that the best bet was to create a system with a relatively uniform power response that would interact with the room to eliminate most “dead spots” and be fun to listen to. Phase response be damned, it’s a red herring. That’s the 901.


And perhaps it's worth adding that a complete and accurate reproduction of the live event is not the self-evident criterion of merit.  I don't go to concerts of symphonic or chamber music for the ultimate auditory experience, but rather to experience the drama and dialogue of live playing. But each seat will have a (slightly?) different balance of sound; some will be distinctly suboptimal. Some really well-made recordings will give an experience of a piece that perhaps could not be heard in *any* seat in any hall. Stereo is good because it helps to hear the different voices in a complex piece: maybe multichannel would be better; but that's the value, not realism.

Michael

"Putting the Science Back into Loudspeakers"

Reply #17 –
Quote from: BradPDX on
...Way, way back in college I was a TA for Dr. Amar Bose (I will pause here until the heckling dies down)...

The biggest problem with Bose is their extremely overpriced systems. So one would expect a lot from them and they don't deliver.
Quote from: BradPDX on
...Though we disagreed on many things, he convinced me of one right away. Simply stated, stereo recording is a sham with no hope of ever sounding like a convincing live event...

Strangely Bose claims this 2.1 system can deliver a surround sound experience!!!

"Putting the Science Back into Loudspeakers"

Reply #18 –
Neither of those points has anything to do with anything Brad said.

"Putting the Science Back into Loudspeakers"

Reply #19 –
Quote
So, given the realities of the products that will be used with speakers (LPs, tape and CDs) he felt that the best bet was to create a system with a relatively uniform power response that would interact with the room to eliminate most “dead spots” and be fun to listen to. Phase response be damned, it’s a red herring. That’s the 901.



The importance of uniform power response (on- and off-axis behavior)  has become a central tenet in speaker design based on NRC work of Floyd Toole et al published in the 80's and onward.  In this respoct Dr. Bose was ahead of his time.  ;>

The thing is, though I've never been able to find any actual MEASUREMENTS of Bose 901 systems, to see how theory has been implemented in reality  (IIRC, Bose went for a 30/70 ratio of direct to reflected sound...I'm not sure how the 901s would be classified. They're certainly not ominpolar, perhaps a variant of bipolar)