Free audio comparison software

[paulrichards7]

http://www.libinst.com/Audio%20DiffMaker.htm
Ive found this software that subtracts one audio file from another to give you the difference. Can be useful in testing cables, capacitors etc, etc
The only thing im not sure of is, if you can figure out if the difference is subtractive or additive?
It has a lot of other preferences that I dont understand so I could be missing something. Im interested to see what you guys think
Ive used to to check out cables and see what my patchbay is doing to the audio signal, but like I said I cant tell if the difference is additive or subtractive
Let me know your discoveries!
Paul

 

[Jonte Knif]

Interesting topic, one of my favorites. And that software looks very handy.

In general you can not tell for sure whether something was subtracted or added if you don't have some extra info like phase. In cables and capacitors there are in principle mostly losses, but then, for example cable capacitance can cause a resonance together with an amplifier stage etc.

Some interesting variants are better done in DAW.
You only need a plugin to swap phase and in some situations a fine gain adjustment is needed. 0,01 dB is enough for most purposes I've encountered. I've done a lot of subtractive comparisons for time-variant things like capacitors (memory effects, interesting results) and to find out power supply quality in amps etc. I'm mainly interested in effects that can not be put simply into THD numbers or freq response graphs. When I build things I use meters only to check that nothing is actually broken   I feel that subtraction test is a handy tool in some situations, but the results get too often masked by phase differences. So you have to know what to listen OR come up with test methods which cancel the phase problems like mentioned on that page. How ever, things are sometimes more complex than they seem.

Any way, if the distortions/losses are really complex in nature, I would say that you can not tell the difference between subtractive and additive without _very accurate_ RMS meter. Normal plugins wont handle the job. On the other hand if there are simple lossy mechanisms, you could do this in DAW: compare the difference waveform to the "original" or better signal chain. If it seems to be mostly in phase, you got subtraction.

One time we compared capacitors. It was kind of fun to notice, that between straight wire and normal polyprop caps there was almost no difference, but one high end tin foil cap was not quite neutral. Electrolytics and tantalum caps were the worst, of course. It is hard to qualify or quantify the effect of these imperfections, but to make them easily audible with subtraction tests is fun.

One interesting thing you can do for certain equipment is to run signal through it second time in anti phase and then add. So in effect you are subtracting, but you can trace interesting phenomena in certain devises. One would assume that the remainder is only odd-order distortions and nothing else, but that is not always the case. Some equipments are "sensitive to phase" even if they shouldn't be, and not only in terms of the phase of even order distortion. For example oxide layer in connectors can act strangely. Try this with your patch.

One time we compared a transformer before and after one week run in period. The samples (6 songs) were also delivered to a couple of people to do blind tests. The subtraction revealed extremely small difference, below -60dB, but some people seemed to hear it. Something had changed, transients seemed to "jump off" from the subtraction. Small differences in core linearity, perhaps?

I hope you can get some sense of my less than perfect explanations and expressions.

-Jonte

 

[abbey road d enfer]

You only need a plugin to swap phase and in some situations a fine gain adjustment is needed. 0,01 dB is enough for most purposes I've encountered.

That would mask any artefact lower than -60dB.
When you want to evaluate the effect of a coupling cap or a contact (patchbay or swich), the artifacts are generally lower than -80dB. In order to make the artefacts stand out, i.e. with less than -86dB residual, you need to balance the phase and antiphase signals with 0.0005dB accuracy.
The substractive test alone is just a way of indicating that there is a difference between the phase and anti-phase signals, but it's very hard to tell if the difference is in the tested signals or in the test jig. Since it is a very sensitive test, any tiny difference in the signal path stands out, in particular phase tracking must be accurate to less than 0.03° all over the audio bandwidth, and frequency response must track to less than 0.0005dB.

 

[JohnRoberts]

As others have mentioned null testing is a kitchen sink test so it captures ALL differences, many of which are insignificant.

That said it is definitive. When you null to -XXX dB all that's left is the sound of the angel's footsteps as they dance on the pinhead... it's hard to hear differences that don't exist. 

JR

 

[Jonte Knif]

As others have mentioned null testing is a kitchen sink test so it captures ALL differences, many of which are insignificant.

Exactly.
It takes time to learn what is important, and I'm not yet there.

When you want to evaluate the effect of a coupling cap or a contact (patchbay or swich), the artifacts are generally lower than -80dB. In order to make the artefacts stand out, i.e. with less than -86dB residual, you need to balance the phase and antiphase signals with 0.0005dB accuracy.

if by coupling cap we mean modern high quality MKP, then that might be so, how ever, if you compare electrolytics and tantalum even -40dB accuracy will take you to levels where you can clearly hear the problems. I think of it as a magnifying glass or microscope.
And signal doesn't get masked by a signal of the same level as we know, so very often 60dB "magnification" is more than enough. If you can not hear very clear problems, you definitely can not hear them 60dB down in level. But I see and understand the "lab" thinking. For me 60dB is enough.

The substractive test alone is just a way of indicating that there is a difference between the phase and anti-phase signals, but it's very hard to tell if the difference is in the tested signals or in the test jig. Since it is a very sensitive test, any tiny difference in the signal path stands out, in particular phase tracking must be accurate to less than 0.03° all over the audio bandwidth, and frequency response must track to less than 0.0005dB.

I understand that at those levels things get tricky, but I didn't quite understand how in this case. The measurement system would constitute of a high quality sound card or converters. They don't leave much residual. Between two music/signal samples the responses surely have not drifted that much. Did I miss something important?

Obviously comparing a gear to wire is just way too fuzzy, it is not going to work.

I'll try to explain one interesting (for me, that is) test. The sample should be percussive and rich, we used tambourine. Make another version of it by cutting some 100ms of the attack away. The "tail" will be your sample 2. First run the whole sample trough the device under test twice and listen that no residual is present. Then run the shorter sample through DUT. Now, produce the difference signal for the "tails". This reveals quite a lot of interesting time variant things. They can be cap dielectric absorbtion , moving tube grids, sagging power supplies etc, you tell me.  It is very hard to examine these things and their contribution to audio with lab instruments. So we are somewhere between full objectivity and subjectivity. Some people find it uncomfortable, for me it is fascinating. And note that after the first phase you can be sure, that the system works. There are no problems associated to freq or phase shifts. Modern well designed gear pass this test with ease, older stuff not. Then you just have to listen to the residual and think if it is pleasant or not, does it mask the resolution, is it muddy, harsh, distorted etc. I've heard them all, I guess : And after that get confused and stay awake all night

-Jonte

 

[JohnRoberts]

I think capacitor DA is overstated as a phenomenon for most (constant termination impedance) applications, but there are many mechanisms associated with drifting or dynamically changing operating points (especially in vacuum tube paths).

Have fun, it (nulling) is a powerful tool, but like any tool it requires some experience to get the most out of it.

JR

 

[abbey road d enfer]

if by coupling cap we mean modern high quality MKP, then that might be so, how ever, if you compare electrolytics and tantalum even -40dB accuracy will take you to levels where you can clearly hear the problems. I think of it as a magnifying glass or microscope.

If you can clearly identify it, that would mean that the capacitor introduces a distortion with an order of magnitude of 1% (probably a typical value of 0.1%). I have never measured that on a working electrolytic; a non-working lytic can be traced easily, being off-value or leaky.

And signal doesn't get masked by a signal of the same level as we know, so very often 60dB "magnification" is more than enough.

That would be true if the artifacts were not correlated, then they would stand out, but most of the time they are, that's why you need the artefact to be at least equal to the residual to be accurately identified.

The substractive test alone is just a way of indicating that there is a difference between the phase and anti-phase signals, but it's very hard to tell if the difference is in the tested signals or in the test jig. Since it is a very sensitive test, any tiny difference in the signal path stands out, in particular phase tracking must be accurate to less than 0.03° all over the audio bandwidth, and frequency response must track to less than 0.0005dB.

I understand that at those levels things get tricky, but I didn't quite understand how in this case. The measurement system would constitute of a high quality sound card or converters. They don't leave much residual. Between two music/signal samples the responses surely have not drifted that much. Did I miss something important?

The accuracy of the conversion process is as good as the stability of the converter's reference voltage. Since it is not a particularly sensitive subject, most of the converters drift somewhat over time and in relation to temperature and mains voltage
essentially. No figures are published in this respect, but I would think the variations are just about what it takes to unbalance the level just enough to ruin the differential method

I'll try to explain one interesting (for me, that is) test. The sample should be percussive and rich, we used tambourine. Make another version of it by cutting some 100ms of the attack away. The "tail" will be your sample 2. First run the whole sample trough the device under test twice and listen that no residual is present. Then run the shorter sample through DUT. Now, produce the difference signal for the "tails". This reveals quite a lot of interesting time variant things. They can be cap dielectric absorbtion , moving tube grids, sagging power supplies etc, you tell me.  It is very hard to examine these things and their contribution to audio with lab instruments. So we are somewhere between full objectivity and subjectivity. Some people find it uncomfortable, for me it is fascinating. And note that after the first phase you can be sure, that the system works. There are no problems associated to freq or phase shifts. Modern well designed gear pass this test with ease, older stuff not. Then you just have to listen to the residual and think if it is pleasant or not, does it mask the resolution, is it muddy, harsh, distorted etc. I've heard them all, I guess :

Then you know that there is a difference, but you don't know which one is good, which one is bad (both probably)

And after that get confused and stay awake all night

No, I'm going to bed 

 

[Jonte Knif]

Test done.

converters were HEDD DA and Burl Bomber AD. (One might argue that Burl is not a good all around instrument for this and we'll see later that this is true)
44,1kHz / 24bit, all operations in Pyramix software.

Music was, well, who cares, pop with fairly big dynamics for modern pop.

First samples recorded right after each other leaved extremely low residual, hard to say how much, buried in the noise but audible .

Then after 4 hours the gain had drifted so much that the residual was at -70dB and sounded _almost_ like the original, but dirty and lifeless. Well, one wasn't listening with too many bits resolution at this point, perhaps around 8-9 bits

Another sample was made within 30 minutes with phase reversed. Here we got -78dB residual, horribly distorted. That reveals the amount of even order distortion in either of the converters, and knowing the rather unorthodox transformer coupled input in Burl I don't hesitate to blaim it. Without the distortion the residual would have been -80dB maximum

So, if you are trying to do null test with anti phase _feeding_ do check first with straight wire that the converters are not distorted. When feeding in phase all distortion will cancel (if the gain has not drifted)

Next time I do null test I'll first start with extensive test with at least twice the lenght of anticipated test time to know the speed of drift. In 4 hours it was significant but not in 30 minutes.

I still stay with my claim about the capacitor test results. It was _not_ about time invariant distortion and we didn't hear any. We heard obvious lack of resolution in time domain and severe pumping effects in the residual, like total dynamics modulating something else. (sounded a bit like modulated filter)

I don't need 0,001 dB resolution to hear things like this, and I know my AP would be of no help measuring the distortion with steady state signals.

I also stay with my opinion about required nulling accuracy for most effects. One can look it in many ways, but for me it is absolutely, definitely convincing that if I can not hear anything weird after 60dB "magnifying" there is nothing to be heard in normal use. This is what null testing is all about for me.

And signal doesn't get masked by a signal of the same level as we know, so very often 60dB "magnification" is more than enough.

That would be true if the artifacts were not correlated, then they would stand out, but most of the time they are, that's why you need the artefact to be at least equal to the residual to be accurately identified.

Really? the most easily buried distortion is 2nd harmonic, and that is easily heard at 3% and positively identified at 10%, at -20dB. Higher order harmonics are easily heard at much lower levels. I don't get it why you should have almost no "original" signal left when evaluating.

About the "tail" test I described:

Then you know that there is a difference, but you don't know which one is good, which one is bad (both probably)

Yes, bad in a way that the DUT is not stable in time domain. But there is only one DUT and if there is residual after very dynamic event, this is all I have to prove. The DUT doesn't handle highly dynamic material. You don't have to compare anything to another DUT, just check that straight wire is OK.   I assume (and can test) that the error is related to dynamic event leaving "memory" somewhere in the circuit. One gear we tested leaved a noise residual over the cymbal tingling in the tambourine. I have no explanation to this, and this is how far the test will get you. If you don't identify the mechanism, you are completely lost if the aim is to improve the device. Harmonic distortion etc are piece of cake to deal with compared to time variant phenomena.

Thank you for pointing up me the drift problem in converters! Valuable info.

peace,

Jonte

 

[abbey road d enfer]

Quote
That would be true if the artifacts were not correlated, then they would stand out, but most of the time they are, that's why you need the artefact to be at least equal to the residual to be accurately identified.

Really? the most easily buried distortion is 2nd harmonic, and that is easily heard at 3% and positively identified at 10%, at -20dB. Higher order harmonics are easily heard at much lower levels. I don't get it why you should have almost no "original" signal left when evaluating.

This is true if your test signal is pure sine wave. On music, it takes another order of magnitude to be noticeable, because there are already a number of harmonics and formants.

 

[Jonte Knif]

Quote:
the most easily buried distortion is 2nd harmonic, and that is easily heard at 3% and positively identified at 10%, at -20dB. Higher order harmonics are easily heard at much lower levels. I don't get it why you should have almost no "original" signal left when evaluating.

This is true if your test signal is pure sine wave. On music, it takes another order of magnitude to be noticeable, because there are already a number of harmonics and formants.

Interesting. So, do I not understand English (quite possible) or are you saying, that you/someone/me could not tell if a sample of music has 10% of 2nd harmonic distortion? I've tested (and I'm not the only one, for sure) this extensively with adjustable 2nd harmonic distortion and depending on material and listening volume the threshold of audibility is (conservatively) somewhere between 0,5% and 3%. With complex material and not too much dynamics like whole mixes the effect is easier to notice because of the amount of intermodulation products, which have nothing to do with any of the harmonic series of single instruments and only get masked by the normal masking effect in ears and not by partials exactly in the same place. (try choir music) One bass note from piano is also a good test signal, because the harmonics are so much out of tune. Harmonic distortion will produce many sets of partials from the partials in the note, and they ALL will be "out of tune" with each other and all the intermodulation tones. 

If you really want to find test signal which makes things difficult to identify, I suggest sawtooth, not music, not even sine wave, although it is a good candidate because of the lack of intermodulation.

A brief but good introduction to masking by pure tones, narrow band noise and broad band noise can be found for example from Rossing's Science of sound. Recommended, although, for obvious reasons music is not included, but rough estimates can be done from the curves presented.
A few examples:
Masking a pure tone by broadband noise: 20dB-30dB difference is needed to mask. This is easily the worst masking signal. There is a good explanation, which I barely understand.
Masking a pure tone by narrow band noise: heavily dependent on level (better look at the curves). At 60dB noise the masking is -50B at approximately where the second harmonic of the noise center frequency would be. A level of 80dB would mask much broader band. Correlates well with other measurements where critical bands are examined.
Then the one related to simple harmonic distortion:
The masking tone is 400Hz. The curves jump up and down a bit where test tone is close enough to cause interference,  but lets take the second harmonic. At 80dB Masking is -38dB, so somewhat over 1% distortion would be needed to be audible. At 60dB masking is only -50dB, so 0,3% distortion would be heard. This is one reason why it is not a good idea to listen at high SPL when trying to hear distortions (except from speakers, of course) This figure probably gives (to the uneducated test person) the absolute minimum level of 2nd harmonic that is audible in perfectly harmonic tone / set of tones / note / what ever. And I feel that training the ear will not make you much better, because this masking effect is so fundamental and requires no central processing.

Music has too many dimensions to be conveniently put into this kind of graphs, but they give some basic info about what to expect. I can not see any results confirming some of the claims presented about masking here in discussion. We have a difference of about 20dB in our opinions/claims/information/knowledge/test results. Some kind of masking phenomenon?

I think I'm getting slowly out of the original topic, but I could not leave the claims undisputed. After all, they DO have influence on the way we see the requirements of null testing.

peace again,

-Jonte

 

[abbey road d enfer]

I'm not talking about psychoacoustic masking, I'm just talking about the simple fact that if you use a violin, or worse a cello, the level of second harmonic is almost as high as the fundamental. If your DUT produces minor levels of H2, it will pass unnoticed.

 

[Jonte Knif]

Ok, I see.

I wrote:

If you really want to find test signal which makes things difficult to identify, I suggest sawtooth, not music, not even sine wave,

I was a bit sarcastic, sorry for that, but on the other hand the basic string motion in violin is sawtooth, lots of harmonics and all in perfect harmonic relation.

I got confused because you talked about music or signals in general, not about the specific cases where only one instrument with perfectly harmonic partials plays monophonic texture (in acoustically dead space.) Strings or perhaps brass will certainly require a lot of distortion to be heard. I absolutely agree. Even 10% could go unnoticed.

I would never choose something that simple as test material for null tests. There was the confusion why I couldn't understand you, sorry.

-Jonte