Prepare audio clips for online comparison?

[Paul R]

I want to put up a small thread with four audio clips of the same sample of music, say 30 seconds worth, but I am unsure of how to do it. The samples are from the same track on the same vinyl LP, but have been recorded differently. 

 

First thing I ran into was level matching. How to do that non-destructively. Normally I would just compress an audio track, but that does not seem like a good thing to do in tis case.

 

Second is everything else that needs to be done to make it a fair comp?

 

Thanks for any pointers!

 

-Paul

 

[sandyk]

8 hours ago, Paul R said:

I want to put up a small thread with four audio clips of the same sample of music, say 30 seconds worth, but I am unsure of how to do it. The samples are from the same track on the same vinyl LP, but have been recorded differently. 

 

First thing I ran into was level matching. How to do that non-destructively. Normally I would just compress an audio track, but that does not seem like a good thing to do in tis case.

 

Second is everything else that needs to be done to make it a fair comp?

 

Thanks for any pointers!

 

-Paul

 

Paul

 Are the output levels that different ?

 Provided that they are within a few dB of each other it shouldn't matter to experienced listeners that much UNLESS the audible differences are quite small.

 You could of course measure the maximum levels of each track with software such as Sound Forge etc.and provide that info for those who use quality attenuators and know the attenuation of each step.

 Any additional processing is likely to make the differences less obvious.

 

 I presume that the tracks have been converted to at least 24/96, with 24/192 becoming more common these days,  although a few still seem to believe that .mp3 is good enough for the LP medium.

 

 Regards

Alex

[Paul R]

1 hour ago, sandyk said:

Paul

 Are the output levels that different ?

 Provided that they are within a few dB of each other it shouldn't matter to experienced listeners that much UNLESS the audible differences are quite small.

 You could of course measure the maximum levels of each track with software such as Sound Forge etc.and provide that info for those who use quality attenuators and know the attenuation of each step.

 Any additional processing is likely to make the differences less obvious.

 

 I presume that the tracks have been converted to at least 24/96, with 24/192 becoming more common these days,  although a few still seem to believe that .mp3 is good enough for the LP medium.

 

 Regards

Alex

 

They are not that different, but despite my best efforts, different enough that loudness would easily be the overriding factor I am sure.  They are at 24/192k. I suspect that even has much as a quarter dB could influence listeners preference. 

 

Normally, I would just equalize them in Final Cut Pro, but I think that might be defeating the purpose. 😉

[pkane2001]

2 minutes ago, Paul R said:

 

They are not that different, but despite my best efforts, different enough that loudness would easily be the overriding factor I am sure.  They are at 24/192k. I suspect that even has much as a quarter dB could influence listeners preference. 

 

Normally, I would just equalize them in Final Cut Pro, but I think that might be defeating the purpose. 😉

 

Just use my DeltaWave software to compare the two clips (caution: beta software!) It'll tell you the difference between the tracks in dB by doing careful nulling of the two tracks. You can then apply this value in Audacity or another software:

 

https://audiophilestyle.com/forums/topic/55878-deltawave-null-testing-audio-comparator-beta/

 

You can measure channels individually, or use L+R combination to determine the overall difference.

[mansr]

Since it's the same music content, it may be sufficient to simply adjust the volume to match RMS levels. If you want to show something like a high-frequency roll-off, matching levels within a narrower frequency band, say 1-3 kHz, might be better.

[Paul R]

56 minutes ago, pkane2001 said:

 

Just use my DeltaWave software to compare the two clips (caution: beta software!) It'll tell you the difference between the tracks in dB by doing careful nulling of the two tracks. You can then apply this value in Audacity or another software:

 

https://audiophilestyle.com/forums/topic/55878-deltawave-null-testing-audio-comparator-beta/

 

You can measure channels individually, or use L+R combination to determine the overall difference.

 

Thanks. I generally try my best to avoid Windows software, but this seems like it should be one of those exceptions.  

 

Appreciate the clear advice too Paul.

 

Yours,

-Paul

 

P.S. Any chance of a Mac version in the future? 

[pkane2001]

16 minutes ago, Paul R said:

P.S. Any chance of a Mac version in the future? 

 

It should be possible, but since I'm using a number of third-party libraries I'll need to see if they'll work on MacOS. 

[sandyk]

10 hours ago, Paul R said:

 They are at 24/192k. I suspect that even has much as a quarter dB could influence listeners preference. 

 

 Agreed, that when the audible differences are quite small, that you need to go to that extent.

[esldude]

If frequency response is really flat for all not much difference doing it peak or RMS.  If there is variation RMS will be what you want.  So use RMS.  I've used Deltawave quite a bit on the recent 8th generation files I posted.  You can trust it to give you differences to set levels by.  It usually agrees with my manual methods in Audacity until the 3rd decimal place.  And it should be the more accurate method, but Audacity works fine.  

 

If there are significant response differences, especially at frequency extremes, you could use what is suggested for speaker matching.  Match levels for the 500-2000 hz range with filters rolling off above and below that point.  2nd order filtering seems sufficient.  

 

Also if using Audacity be careful of it being set for dither.  If you aren't changing sample rates and bit depths, leave dither set to none.  So open in Audacity and use amplify to adjust all the files to the level of whatever the reference is.  Save the result.   If you can use Windows Deltawave does all this for you including allowing you to play while doing an ABX comparison.  

[sandyk]

1 hour ago, esldude said:

If frequency response is really flat for all not much difference doing it peak or RMS.  If there is variation RMS will be what you want.  So use RMS.  I've used Deltawave quite a bit on the recent 8th generation files I posted.  You can trust it to give you differences to set levels by.  It usually agrees with my manual methods in Audacity until the 3rd decimal place.  And it should be the more accurate method, but Audacity works fine.  

 

If there are significant response differences, especially at frequency extremes, you could use what is suggested for speaker matching.  Match levels for the 500-2000 hz range with filters rolling off above and below that point.  2nd order filtering seems sufficient.  

 

Also if using Audacity be careful of it being set for dither.  If you aren't changing sample rates and bit depths, leave dither set to none.  So open in Audacity and use amplify to adjust all the files to the level of whatever the reference is.  Save the result.   If you can use Windows Deltawave does all this for you including allowing you to play while doing an ABX comparison.  

 I don't agree with that.

If you are trying to demonstrate small differences between different methods, you should ONLY need to set the peak level, nothing more, as most people prefer the louder sounding file. Otherwise you may end up obscuring the small lower level differences that you are trying to demonstrate, or perhaps when demonstrating apparent differences in Dynamic range. 

[mansr]

7 hours ago, esldude said:

If frequency response is really flat for all not much difference doing it peak or RMS.

Peak level is unreliable. A sub-sample shift can cause a change of several dB in peak level. RMS doesn't have this problem.

[esldude]

2 hours ago, mansr said:

Peak level is unreliable. A sub-sample shift can cause a change of several dB in peak level. RMS doesn't have this problem.

I had in mind working with digital files where I would embed a reference tone. That works as long as you don't use a tone that's an even multiple of the sample rate. Can't do that with analog sources.

 

The proper way is to use RMS and avoid the issues.

[Paul R]

8 hours ago, esldude said:

I had in mind working with digital files where I would embed a reference tone. That works as long as you don't use a tone that's an even multiple of the sample rate. Can't do that with analog sources.

 

The proper way is to use RMS and avoid the issues.

 

Yes, I agree. The files, at least to me, were so different in so many ways - well - I done my best to get them fairly equal. When you see what they are, and especially if you through them under analysis, I think the differences will be even more blatant than just listening. And to me, the listening is quite blatant enough! 

 

Part of this is based upon suspect advice from a vinyl "archivist."  Some of the stuff he said didn't sound kosher, so I wanted to try it myself.  Hoo boy, Feel free to tell me I 'effed up by the way. Appreciate the help folks! 

 

Files are posted at 

 

 

Yours,

-Paul 

[John Dyson]

On 3/29/2019 at 7:39 AM, Paul R said:

 

They are not that different, but despite my best efforts, different enough that loudness would easily be the overriding factor I am sure.  They are at 24/192k. I suspect that even has much as a quarter dB could influence listeners preference. 

 

Normally, I would just equalize them in Final Cut Pro, but I think that might be defeating the purpose. 😉

About the 1/4 dB difference thing.  When doing my testing, it has been amazing to me that a slight difference in frequeny response (not far off from you 1/4dB) can make lightyears difference in how material can sound.  When doing EQ for my mastering for my own listening, 1.5dB is on the high side for EQ at HF, and about 3dB for the LF.  Any more than that, it is usually destructive.  Also, unless I want to make MAJOR differences in the sound, I keep the Q of the filters between 0.50 and 0.707 for shelving, and below 1.0 for peaking EQ.   But, as you write, a difference of 1/4dB can make a substantive difference.  (For EQ OR level.)

 

If you want to compress -- keep it mild and keep the gain changes shallow.  It is best to avoid compressing on the peaks (but compress in the middle signal levels), and only limit on the peaks.  Keep the limiting well below 1.0dB if you can.  My compressor has totally dynamic attack/decay (10msec A/D to over 3sec A/D -- either one) based upon the signal stats.  It is mostly inaudible, but if using something else -- RMS tends to sound smoother, but the FET type of compressors tend to sound more stressed (sometimes desirable) at larger amounts of compression.  The optos are a good tradeoff -- RMS does have its problems unless implemented very, very carefully.

 

Add-on:  for attack/decay time if not dynamic -- reasonably fast so that inter-syllable pumping isni't as noticeable, and low compression ratio.  A long release time is often not a good thing unless the slow gain-up sounds good to you.  Multi-band is your friend, but multi-band at a high compression ratio and large gain change is multi-bad.

 

These differences are by MY perception, and I am NOT a golden ears...  My hearing is very flawed and I can easily detect the differences.  Of course YMMV.

 

John