Does Phase Distortion/Shift Matter in Audio? (no*)

[j_j]

Thank you for your response, @j_j

Conclusion. After all this reading, it seems to me that not only is phase audible, lack of phase distortion seems associated with better spatial characteristics. As a medical scientist, I would say that this does not even reach Level C evidence (i.e. the published evidence is very weak, no consensus between experts).

I agree with your analysis. The question of "is intra-aural phase audible" depends massively, and I do mean MASSIVELY on the kind of phase shift. I regard points 2, 3 and 4 as mostly equivocation (although they are right, they argue to usefulness in the present day, not to actual audibility).

As to point 1, I can easily provide you with an example wherein phase shift in a pair of signals with absolutely identical (to double precision) power spectra sound very different. Now, this is a "nearly absurd" level of phase shift, indeed, but it is an end-point that simply can not be controverted. Those signals are actually ON this site somewhere, I forget where, in a ??different thread maybe??.

If you have matlab available, I'll simply copy a script to generate them in a following article. (I have to go find it, and the "proof" of exact power spectrum will be very obvious. I may even rewrite the script to provide less radical phase shifts, if you want to play with it.

 

[gberchin]

Hmmm. I'd have to look. Is it in phase at all frequencies?

As I recall, LR crossovers are considered to be in-phase at all frequencies, but in truth they are multiples of 2pi radians apart, depending upon the order. That's why their step response is so ugly.

In matched-delay subtractive crossovers based upon Gaussian or Bessel filters the lowpass and highpass sections are in-phase with each other, though the Bessel responses start to diverge at some point well above the crossover frequency.

In matched-delay subtractive crossovers based upon Butterworth or Butterworth-squared filters, the lowpass and highpass sections are never in-phase with each other.

 

[gberchin]

Consider. Narrow bandwidth ::= long impulse response.

I understand. However, duration of the pre-ring also depends upon the crossover frequency (lower frequency = longer duration), so at some point does temporal masking become the dominant factor?

 

[Keith_W]

If you have matlab available, I'll simply copy a script to generate them in a following article. (I have to go find it, and the "proof" of exact power spectrum will be very obvious. I may even rewrite the script to provide less radical phase shifts, if you want to play with it.

Thank you JJ, but remember who you are talking to. Medical scientist. Not an engineer. High school level understanding of maths. And no Matlab! My abilities are very limited.

Regardless, what I am interested in are studies, not demonstrating to myself if it is audible or not. I already think it is audible. But this is anecdote and not evidence. Nobody is convinced by anecdote, at least good scientists should not be. If I went around ASR loudly proclaiming that Toole is wrong and we can hear phase shift because I heard it, I would rightly be ridiculed because I have no evidence.

 

[j_j]

clc
clear all
close all

fs=44100;

f=500;

del=7;

len=2^17;

ifreq=(round(f/fs*len)) + 1
idel=round(del/fs*len)


xt(1:len,1)=0;
xt(ifreq)=1;
xt(ifreq-idel)=.5;
xt(ifreq+idel)=.5;

xt(len:-1len/2+2))=conj(xt(2len/2)));

x=ifft(xt);

am=x;

plot(am)

xt(1:len,1)=0;
xt(ifreq)=1;
xt(ifreq-idel)=.5;
xt(ifreq+idel)=-.5;

xt(len:-1len/2+2))=conj(xt(2len/2)));

x=ifft(xt);
fm=x;

hold
plot(fm)

t=max(max(abs(am)),max(abs(fm)));

am=am/t*.5;
fm=fm/t*.5;

audiowrite('am.wav',am,fs,'BitsPerSample',16);
audiowrite('fm.wav',fm,fs,'BitsPerSample',16);

 

[Keith_W]

I understand. However, duration of the pre-ring also depends upon the crossover frequency (lower frequency = longer duration), so at some point does temporal masking become the dominant factor?

Read this, specifically section 2 on "temporal masking".

The reference is Zwicker and Fastl.

 

[j_j]

Thank you JJ, but remember who you are talking to. Medical scientist. Not an engineer. High school level understanding of maths. And no Matlab! My abilities are very limited.

Regardless, what I am interested in are studies, not demonstrating to myself if it is audible or not. I already think it is audible. But this is anecdote and not evidence. Nobody is convinced by anecdote, at least good scientists should not be. If I went around ASR loudly proclaiming that Toole is wrong and we can hear phase shift because I heard it, I would rightly be ridiculed because I have no evidence.

To point out what matlab provides, it makes two files with identical power spectra. They can be trivially played in an ABX device. I am pretty sure that 100% of people with functional hearing will be able to pass the ABX test with nearly 100% results. That's certainly been true of everyone I've showed this to.

I don't have time to run a 100 person 20 trial test. Maybe you know somebody who could. Then they could write the paper and end the argument.

Until then, all I offer is an absolute controversion to the claim "phase is not audible". As a hypotheses, it fails hard.

Any good scientist must respect a clear falsification.

 

[j_j]

Read this, specifically section 2 on "temporal masking".

View attachment 399618
The reference is Zwicker and Fastl.

It is very, very out of date, just FYI.

 

[j_j]

I understand. However, duration of the pre-ring also depends upon the crossover frequency (lower frequency = longer duration), so at some point does temporal masking become the dominant factor?

Consider the other half of the question, "how steep does the filter have to be". Cochlear filters are very steep. How often would you need a very, very short transition band (i.e. long filter)?

Good news is "you don't".

 

[Keith_W]

It is very, very out of date, just FYI.

The book, or the evidence? Is there newer evidence with different thresholds on pre-masking?

 

[j_j]

The book, or the evidence? Is there newer evidence with different thresholds on pre-masking?

I am not aware that anyone has written accurate (or even inaccurate) work on pre-masking vs. frequency content. It is time. I am no longer paid to do this kind of research, but it is easy to construct such a test in the modern day.

What is missing is information on frequency content vs. premasking time. There is some alluded to in Johnston and Brandenburg (or vice versa)'s paper on Hybrid Coding, from many, many years ago. Yes, I wrote much of that paper, no, I don't recall if there is a specific cite.

https://secure.aes.org/forum/pubs/conventions/?elib=5757

Yah, it's been a long, long time.

 

[gberchin]

The reference is Zwicker and Fastl.

Yes, thank you. I read it from cover to cover in ... oh, I believe it was around 1992. So the info is familiar to me, but it's been a long time.

 

[gberchin]

How often would you need a very, very short transition band (i.e. long filter)?

Good news is "you don't".

I'm currently building a set of speakers with horn-loaded HF section. Anything less than a 4th-order transition band is not recommended, since compression drivers unload quickly below the cutoff frequency of the horn, and even more below their resonance frequency.

 

[j_j]

I'm currently building a set of speakers with horn-loaded HF section. Anything less than a 4th-order transition band is not recommended, since compression drivers unload quickly below the cutoff frequency of the horn, and even more below their resonance frequency.

This is one of the reasons an FIR crossover is a big advantage, actually.

 

[witwald]

I recently heard a demo (at the Melbourne hi-fi show) of a linear phase loudspeaker that made me think about this subject of intrachannel phase distortion a little more. The speaker was set up very close to the listeners, I was about 1m (3ft) away from the speakers. There was exceptional clarity to the sound. Why?

One reason might be that the speakers in question might have had a non-flat frequency response. This could easily give the impression of better "clarity". Near-field monitoring, as the speakers were set up very close to the listeners, also has its own "clarifying" effect, by reducing the influence of room reflections at the listening position.

Firstly, loudspeakers are minimum phase devices that by themselves produce phase distortion.

Although individual loudspeaker drivers are generally minimum-phase devices, in the absence of first-order crossovers, multi-driver loudspeaker systems do not possess minimum-phase characteristics.

Secondly, room reflections produce excess phase and add to that phase distortion.

Doesn't the excess phase first come from the flight time from the driver/loudspeaker to the listener? Room reflections would not generally be regarded as contributing to excess phase.

By the sound reaches your ears, there is so much phase distortion that it swamps any additional phase distortion that may be introduced by electronics.

In general, phase distortion in electronics is going to be quite minimal. The worst-case scenario is probably going to occur if the electronics are band-limited to cover only the 20Hz to 20kHz frequency range. Today, that would be regarded as an unusual and limiting design choice. For example, many amplifiers are DC coupled and have −3dB high-frequency cut-off points that are at least 80kHz, and often more. If they are not DC coupled, then they can have a −3dB low-frequency cut-off frequency of 5Hz or less.

 

[René - Acculution.com]

Not an expert on the topic at all, but would say I have heard a difference in a very short blind test, with particular track, so very anecdotal. But I think an issue with earlier studies was that allpass filters were added to normal loudspeakers that in themselves already had phase "issues" (not linear phase), and then people were asked "is this the same or not?", whereas with recent advanced is easier to phase linearize a loudspeaker and ask the same question.

 

[antcollinet]

But how do you know that that is true?

You can test it with a technical question you know the answer to. For example
"What benefits can I expect from expensive speaker cable"

Expensive speaker cables can offer several potential benefits, though the degree of improvement often depends on your audio system and personal preferences. Here are some advantages you might expect:

1. Improved Signal Transmission: Higher-quality cables often have better conductivity, which can lead to more accurate sound reproduction and reduced signal loss.
2. Enhanced Build Quality: Expensive cables are typically constructed with superior materials, making them more durable and resistant to wear, which can be particularly beneficial in high-traffic areas.
3. Better Insulation and Shielding: Quality cables often feature advanced insulation materials that reduce interference and electromagnetic noise, helping maintain audio clarity.
4. Reduced Resistance: Lower resistance in premium cables can result in better power transfer from your amplifier to your speakers, potentially improving dynamics and overall sound quality.
5. Improved Connectors: High-end cables often come with better connectors that ensure a more secure and stable connection, minimizing the risk of signal loss.
6. Aesthetic Considerations: Expensive cables can have a more polished appearance, which some users appreciate for the overall look of their audio setup.
7. Brand Trust and Warranty: Established brands may offer better customer support and warranties, providing peace of mind with your investment.

While many users report noticeable improvements, others may find minimal differences. It’s often best to try cables in your system to see if you perceive any benefits.

The answer is steeped in audiophile folklore. All the stuff about improved sound quality, known to be no different between expensive speaker cables and £1/ft zip wire. And just like that no-nothing uncle I mentioned above, it mixes in just enough stuff that seems to make sense (eg "improved signal transmission" - even though spending more doesn't buy that at all) to make it seem as though the answer is good.

The last statement is particularly telling - in fact the whole answer completely fails to even mention the reason people hear differences in well contsructed cables at any price. (percpetual bias)

In other words gpt4 has been trained on - and regurgitates - all the sort of stuff you can read all over the audio based internet - regardless of its technical veracity.

If it is so flawed for something as simple as speaker cable - how can you believe it is any better for more complex subjects.

Then the answers it gave that you posted above look like regurgitations of company sales literature. If part of the training set includes sales blurb - then clearly it is totally untrustworthy.

How do you that what that ARS post says is true???

(I assume you meant ASR).
At least with human information sources, you can form an opinion of the reliability of the source - based on what you can learn about their technical capability, and motivation to be truthful. So high credibility news sources for example risk reputaitonal loss when they geet stuff wrong and so on.

Also based on how accurate they are on stuff you already know about. (As I demonstrated above with gpt4)

Once you have identified reliable sources, then fact checking becomes less critical.

Not possible with AI's because you have no idea of the training set, and therefore what comes out is a total crap shoot regarding reliability - even if they have got stuff right 10 times in the past, the 11th time is no more likely to be correct, if the training data in that area is flawed.

 

[j_j]

You can test it with a technical question you know the ansower to. For example
"What benefits can I expect from expensive speaker cable"

...

Not possible with AI's because you have no idea of the training set, and therefore what comes out is a total crap shoot regarding reliability - even if they have got stuff right 10 times in the past, the 11th time is no more likely to be correct, if the training data in that area is flawed.

Yep. And many parts of "AI" are steeped in plagiarized popular culture or such nonsense, to start with. There's no "skepticism" and no "evaluation" of the tommyrot that goes into this stuff.

 

[BDWoody]

Cut and paste of accurate info beats BS from humans

That's enough with the AI stories you are copying and pasting. We have actual experts here to answer questions, and this really isn't what we are looking for.

 

[Holmz]

I think that frequency-dependent phase shift has an…

The technical term for “frequency dependant phase” is “Group Delay”.

….
(I assume you meant ASR).
‘’’

Or he is calling LLM-AIs… arses

That's enough with the AI stories you are copying and pasting. We have actual experts here to answer questions, and this really isn't what we are looking for here.

It was largely accurate, it just reeked of a sales pitch or AI.