PlayClassics test files to compare file formats

[徐中銳]

The picture in your post with the Jared Sack's comment shows some Grimm Audio LS1 speakers which use Hypex amps. Yet despite a sort of processing in those, DSD comes thru flowing as music rather than as PCM blocks. Laughable is what that is.

Alright, let's open the Thread up ? Readers won't be offended if say other recording label's personnel speaks of their methods and free file samples, say, bmoura who Posted the original photo of :

 

And Cookie Marenco too is sometimes in the Forums ? Yes, just found her cookiemarenco

 

And, in the beginning, Post #1 :

These test files have been directly processed from the 24bit 96kHz master on the protools session. The volume is exactly the same for all of them.

[kumakuma]

Alright' date=' let's open the Thread up ? Readers won't be offended if say other recording label's personnel speaks of their methods and free file samples, say, bmoura ... And Cookie Marenco too is sometimes in the Forums ?

 

I'm sure no one will mind. The more, the merrier.

[sandyk]

I'm sure no one will mind. The more, the merrier.

 

The title of the thread is STILL "PlayClassics test files to compare file formats"

It's NOT about test files from Joe Blogs - Recording Engineer!

It should be up to the original poster (Mario Martinez) as to whether he would like to see the thread further opened up, or not. NOT someone who has been accusing Mario of commercialism and continues to disrupt this thread at every opportunity.

[徐中銳]

Joe Blogs - Recording Engineer!

 

Correction, because... Let's acknowledge tailspn (Tom Caulfield) - Recording Engineer !

 

As evidenced by 2 Birthday DSD Surround Sound Picks from Tailspn

[sandyk]

Correction' date=' because... Let's acknowledge tailspn (Tom Caulfield) - Recording Engineer !

 

As evidenced by 2 Birthday DSD Surround Sound Picks from Tailspn

 

How about you acknowledge the title of this thread instead of trying to disrupt it all the time ?

[Audiophile Neuroscience]

The title of the thread is STILL "PlayClassics test files to compare file formats"

It's NOT about test files from Joe Blogs - Recording Engineer!

It should be up to the original poster (Mario Martinez) as to whether he would like to see the thread further opened up, or not. NOT someone who has been accusing Mario of commercialism and continues to disrupt this thread at every opportunity.

 

I agree with Alex.Others can open their own threads. I have no problem if anyone profits from their genuine endeavors. If I like the music or technique, I stand to benefit.

 

Mario would you PM me the link and code?

 

So, if I understood correctly, everything has been recorded exactly the same way and all at 24@96. The only difference is that some bits have been downsampled?

 

David

[徐中銳]

As he said :

For the purpose of this thread, I have made one track containing about 16 minutes of music. Piano solo, soprano and piano, tenor and piano and flamenco voice and guitar...

 

I have uploaded 1 zip file to the server containing this one track in 7 different formats:

flac 24bit 96kHz

flac 24bit 48kHz

flac 24bit 44kHz

flac 16bit 96kHz

flac 16bit 48kHz

flac 16bit 44kHz

lame mp3 320kbps 48kHz

 

These test files have been directly processed from the 24bit 96kHz master on the protools session...

And as was said after Alex's signature :

At the time, I was more curious if emailing audio and broadband would destroy bits of information. We decided that the uncompressed 96 zip was the least damaging (though almost not existent, there is a slight degradation, not as much as FLAC).

 

Those of us in mastering and recording are dealing with issues of loss of bits during storage and archiving daily...

[Mario Martinez]

So, if I understood correctly, everything has been recorded exactly the same way and all at 24@96. The only difference is that some bits have been downsampled?

 

Yes, all the files come from the same 24/96 master file and they all have the same volume. The only different thing these files have is the downsampling itself.

[Sal1950]

I agree with Alex.Others can open their own threads. I have no problem if anyone profits from their genuine endeavors. If I like the music or technique, I stand to benefit.

 

Mario would you PM me the link and code?

 

David

 

+1 We already asked Mario to move this thread from the "Stinking High Rez" thread so we would have a private place to discuss his files. Wihelm, Please stop disrupting the conversation on this thread. You are more than welcome to open a new thread for files from Cookie or any other producer you'd like.

[Mario Martinez]

The title of the thread is STILL "PlayClassics test files to compare file formats"

It's NOT about test files from Joe Blogs - Recording Engineer!

It should be up to the original poster (Mario Martinez) as to whether he would like to see the thread further opened up, or not. NOT someone who has been accusing Mario of commercialism and continues to disrupt this thread at every opportunity.

 

I agree with Alex.Others can open their own threads. I have no problem if anyone profits from their genuine endeavors. If I like the music or technique, I stand to benefit.

 

+1 We already asked Mario to move this thread from the "Stinking High Rez" thread so we would have a private place to discuss his files. Wihelm, Please stop disrupting the conversation on this thread. You are more than welcome to open a new thread for files from Cookie or any other producer you'd like.

 

I do think it would be best to keep this thread on topic. Thank you.

[徐中銳]

disrupting

 

Frankly, what are your, David​'s and Alex's logic here ?

 

This is, essentially, not peer review science ? As amiable as Mario is, is his methodology and consequences above challenge ?

 

Good day to you outside New Zealand, I'm gonna prepare for some sleep

[JSWolf]

Last night I did some comparisons. I compared the 24/96 master version to all the 16-bit versions and the 24/96 definitely sounds better.

 

The 320k MP3 was awful. It was dry and lifeless. All the fine tonality was missing. The MP3 format just sucked the life out of the music.

 

The 16-bit 48k version was a better than the 16-bit 44.1k version, but just but just a bit better.

 

The 16/96 version was an interesting one. While the 24/96 again is better, the 16/96 is much closer than I thought it would be.One thing I did notice with the 24/96 vs the 16/96 was that the imaging was better for the 24/96 the tonality of the 16/96 was better than any of the other 16-bit versions.

 

I'm not ure if I'll have time to compare the other 24-bit versions, but I should be able to on Tuesday if not today.

[Le Concombre Masqué]

took the test on heavily transformed files (turned to +/- 3dB RX4 eQed wav files processed to SDM (DSD128) via HQPlayer) for it's the way I listen to PCM hence the real life worth taking into account situation as far as I'm concerned.

 

A little later on I played Public Image ripped from a SACD and I had to go retrieve and play my 45 vinyl original : I often think digital is a joke (and vice versa some other times, especially with piano...)

 

Results : 3 things surprised me :

 

the biggest difference was between mp3 and the rest : so recessed ! not harsh or what have you that would mark a difference with "analog", recessed

 

then I was pretty amazed by the 16/44 to 16/48 difference ; might be the first time I was given the opportunity : echoes, reflections made much more sense, integrated much better. All differences between the files seemed minor in comparison

 

Last but not least, I found 16/44 to be a relevant sweet spot not so badly chosen decades ago. 24/96 is better in the sense it provides more matter stuff density what have you to carve a soundstage, thereness etc but I also had to lower a crank the volume vs 16/44 hence it doesn't not translate in terms of more analog or less harsh vs 16/44 and if one is looking for hirez superiority there one could easily be convinced that RB is nice enough.

 

Conclusion : Hirez rules but is no solution to harshness if one has such issues, rather it rules by offering.... more (density)

[semente]

Okay, have had some time to spend on this under reasonably good conditions.

 

I listened to only the 24 96 and the 44 16. MP3 is known not to be fully transparent so I had no interest in it.

 

After a few listens I thought there was a difference at times. Especially in the decay in the hall of the piano. When I would go directly back and forth the difference wasn't really there. One afternoon when doing other things I put these two into random play mode so I wouldn't know at any given time which was playing. No differences were apparent to note.

 

So I ran these through an ABX test. 16 trials done twice. 10 of 16 and 9 of 16 done a couple hours apart. Had I gotten positive results or heard differences in the sighted part of the test with conviction I would have tried some of the in between rates and bit depths.

 

After all listening I upsampled the 4416 file to 2496. Dumped it into a sound editor. Flipped polarity and mixed it with the original 2496 file. You get noise that wiggles slightly around the -88 or 89 db level. What you would expect from 16 bit dithering. I boosted this by 40 db, and listened. You hear faint swishing noise. There are some points in the latter portion of the track where you get low level clicks. They coincide with places the original clipped or very nearly did. I imagine the up and down sampling may have caused some intersample overs to show up this way. I then took this file and slowed it down by a factor of 4. This would make it easy to listen to any ultrasonic effects one normally doesn't hear. This was still boosted by 40 db mind you. Really nothing to report, a bit higher pitched tone to the swishing hiss much like FM radio hiss between channels.

 

Here is an FFT of one of the louder (though unclipped) sections using this mix between original and upsampled 4416. Note this is not boosted in level.

 

[ATTACH=CONFIG]22533[/ATTACH]

 

You can read the RMS level for the 0-40 khz band is -89 db. I used shaped dither in the upsampling which is what gives the noise floor the shape you see.

 

I don't see any reason to think these would be audibly different.

 

Hi Dennis,

 

Would it possible for you to upload your null test?

I think that it might be interesting to listen to it.

 

R

[esldude]

Hi Dennis,

 

Would it possible for you to upload your null test?

I think that it might be interesting to listen to it.

 

R

 

Well I could. I don't know how interesting it will be. By itself at normal volumes you will hear nothing. With enough gain you will hear hiss. I'll look into it when I get back home.

[esldude]

Okay, here is 30 seconds from the 14:30 to the 15:00 parts of the file. I chose this portion because it has the most difference between the two versions. I suspect that might be because the voice was nearly overloading the microphone at times. I could be wrong about that however.

 

The path was the 16 bit 44.1 khz version of Mario's recordings which I upsampled to 24 bit 96 khz. I then inverted this upsampled file and mixed it with the unprocessed 24 bit 96 khz original. You would expect to have any difference above 20 khz and dither remaining. You would expect otherwise everything would cancel out to nothing. Also note the dither is the cost of changing the sample rate. Upon playback at native rates at the analog out of your DAC there should be less difference below 20 khz than what shows up in these files.

 

Also doing the same procedure with the 44 24 file all the residuals left over after mixing drop an additional 40 decibels or so.

 

Download the attachment and it should unzip into a 2496 FLAC file.

30 seconds 4416 nulled against 2496.zip

[semente]

Thanks.

I'll try to "examine" the file tomorrow.

 

R

[gmgraves]

...I just sent them to you, Thanks!

 

 

I too would like to hear the Flamenco files Mario. 24/96 FLAC, if possible. Thanks

[Le Concombre Masqué]

since I'm deaf and have a lousy system I don't trust my ears but my eyes

Okay, here is 30 seconds from the 14:30 to the 15:00 parts of the file. I chose this portion because it has the most difference between the two versions. I suspect that might be because the voice was nearly overloading the microphone at times. I could be wrong about that however.

 

The path was the 16 bit 44.1 khz version of Mario's recordings which I upsampled to 24 bit 96 khz. I then inverted this upsampled file and mixed it with the unprocessed 24 bit 96 khz original. You would expect to have any difference above 20 khz and dither remaining. You would expect otherwise everything would cancel out to nothing. Also note the dither is the cost of changing the sample rate. Upon playback at native rates at the analog out of your DAC there should be less difference below 20 khz than what shows up in these files.

 

Also doing the same procedure with the 44 24 file all the residuals left over after mixing drop an additional 40 decibels or so.

 

Download the attachment and it should unzip into a 2496 FLAC file.

[semente]

I haven't had time to listen to the null test but I have run it through spectrum and level analysers.

I have increased the image contrast of the spectrogram between 0.15 and 0.20s.

 

 

 

R

[Alaric. H]

May I have a link to the files when you get a chance thanks.

[Jud]

Well, I tried combining the left channel from the 24/96 track with the right channel from the 16/44.1 track. However, as I should have expected but hadn't thought of, Audacity won't put out a file where the left channel and the right have different sample rates, at least as I know how to use Audacity. It chose 24/96 for the mixed file, either because that was the higher rate or because that was the "top" track in the Audacity UI and that's the one it takes its settings from. This makes sense, too, because I can't imagine a consumer DAC knowing what to do with a mixed sample rate file. What might be needed for a correct comparison is two identical signal chains, one playing back the left channel with one sample rate, one playing back the right channel with another sample rate, unless someone knows how to make one consumer DAC play a mixed file.

 

I also think the starting point does need to be mono so each channel is identical except for any differences due to sample rate. When the original tracks are stereo I believe it would be too difficult mentally to sort out what's due to sample rate and what's due to the other differences between the signals.

[esldude]

Well, I tried combining the left channel from the 24/96 track with the right channel from the 16/44.1 track. However, as I should have expected but hadn't thought of, Audacity won't put out a file where the left channel and the right have different sample rates, at least as I know how to use Audacity. It chose 24/96 for the mixed file, either because that was the higher rate or because that was the "top" track in the Audacity UI and that's the one it takes its settings from. This makes sense, too, because I can't imagine a consumer DAC knowing what to do with a mixed sample rate file. What might be needed for a correct comparison is two identical signal chains, one playing back the left channel with one sample rate, one playing back the right channel with another sample rate, unless someone knows how to make one consumer DAC play a mixed file.

 

I also think the starting point does need to be mono so each channel is identical except for any differences due to sample rate. When the original tracks are stereo I believe it would be too difficult mentally to sort out what's due to sample rate and what's due to the other differences between the signals.

 

Yes, what Audacity does is a real-time resampling. I do believe it defaults to the higher rate. I also am sure you will tell differences better by mono methods.

 

At least as a first step I don't see a problem upsampling the 44 file to 96. The degradation is minimal. Not zero, but awfully low in level. You still have one channel that was subject to bandwidth restriction via brickwall filtering it simply is being reproduced at a higher rate.

 

To resample with Audacity open the 44 file. In the lower left corner change the project rate to 96. Export (using a different name so as not to overwrite the original file). You can then open both the upsampled and 96 file and put them together as you wish.

 

I would look into Preferences, and then under quality. You want your Sample Rate converter set to Best Quality, and you should choose either triangle dither or shaped dither.

 

Do this grabbing say the two left channels. Make one a left and one a right. See if you notice any anomalies. Compare with a file where you used the original 96 left channel in both sides and see if they are different.

[esldude]

Just noticed the file I uploaded yesterday was somehow a 48 khz version. Not sure how I made that mistake. In any case here is the correct one. 16 seconds where I nulled an upsampled 44.1 16 bit file against the original 96 24 bit file. Had to shorten it to 16 seconds to get file size low enough to attach.

null of 96 vs 44 16 16 seconds.zip

[esldude]

And as a bonus, here is the same file where I slowed it down by a factor of 4. That means you can hear all the ultrasonic differences left over between 44 khz and 96 khz. Everything that was between 20 and 40 khz is now between 5 khz and 10 khz.

null 96 vs 44 15 seconds at one quarter speed.zip