Transcoding listening test - Is it as bad as people say it is?

[chewy4]

So I hear a lot of talk about transcoding around here. People say lossy to lossy conversion is the work of the devil. Now the logic behind it is sound, you could be compounding artifacts on top of each other, but is it audible?
 
I found a song with some decent high frequency content. "Veronika's Dream" by Floex. Here's the spectogram:
 

 
I marked the clip I took of it in red, and circled a part that I thought should be particularly susceptible to artifacts.
 
I took a wav of this file and converted to 320kbps mp3. Then I converted it to 320kbps AAC. Then back to mp3. I did this a bunch of times. Probably over 30. Then I converted back to wav. Then back to mp3. Did this about 10 times.
 
Here is the download for the source: http://www.mediafire.com/?rk2d4s0edb235dc

Here are the 320kbps encoded files. One is a plain old 320kbps file. Another is transcoded between aac and mp3 a bunch. A third is that second file but converted back and forth between wav a bunch. NOTE: These files are not volume matched. dbpoweramp is not cooperating with me. If you would like to claim you can hear a difference, please provide an ABX log WITH replaygain enabled.
 
 

Spoiler: 320kbps%20transcoding

 
Clean(only converted from wav): http://www.mediafire.com/download.php?opc9md3239h4amt
Dirty(transcoded several times between mp3 and aac): http://www.mediafire.com/download.php?rkfjwpprlbyu7tq
Extra dirty(second file with extra conversions between mp3 and wav): http://www.mediafire.com/download.php?5e2ywtbxkh3etpa
 

 
 
 
Now I don't know about you guys, but that's not too bad. But don't get me wrong, there is an effect. It is clearly shown in the spectograms.
 
 
Now this next part is particularly interesting. I tried the same with a 128kbps file just to display how transcoding is nasty. I recall doing it with another file just between wav and mp3 and it was ugly. But as I transcoded a good 20 or so times, I noticed there was no terrible distortion. Sure, it was there, but not like I remembered.  So then I tried it between mp3 and wav, and it got nasty fast. Only after a few conversions some blatant ugly distortion came in. If somebody could explain this to me, that would be great. Why on earth would encoding between mp3 and wav repeatedly be worse than between mp3 and aac?
 
Anyways, here are the files. This one should be easy:
 
 

Spoiler: 128kbps%20transcoding

 
Clean(only converted once from wav): http://www.mediafire.com/download.php?g25byxzbtn8rat6
Dirty(transcoded several times between mp3 and aac): http://www.mediafire.com/download.php?xym187byqb69nr9
Extra dirty(second file with extra conversions between mp3 and wav): http://www.mediafire.com/download.php?5x12lbbz0kts69c
 

 
Happy listening! Feel free to share your thoughts.

 

[chewy4]

More spectograms!
 
Top row is 320kbps, from best to worst.
Bottom row is 128kbps, from best to worst.
 
 

 
Higher resolution download: http://www.mediafire.com/view/?hwlyje0a802ce8u
 
 
Judging by these, converting between mp3 and wav seems to add additional data(at least at 128kbps), rather than remove it like compression and transcoding does.

 

[bigshot]

I encoded a file ten times from AAC 256 VBR to AIFF and back again. It sounded fine at the end. AAC is a very good codec. I don't think transcoding is much of an issue. As long as you get to audible transparency you're fine. With my tests, that was AAC 256 VBR. Completely transparent.
 
By the way, when I discovered this, I invited folks here to try it for themselves... particularly those who believe that FLAC is the only way to go. To my knowledge, no one bothered to try it.
 
Try WAV and AAC and see what you find.

 

[chewy4]

I will try wav and 128kbps aac tomorrow night, and I'll see if it gets that distortion that 128kbps mp3 and wav get. I really don't understand where it's coming from, or how mp3>aac avoids it(doesn't this require decompressing to PCM first anyways?). 

 

[bigshot]

I know 128 artifacts. I'd be interested in seeing what happens to files that are audibly transparent... 256 VBR or 320

 

[chewy4]

Quote:

I know 128 artifacts. I'd be interested in seeing what happens to files that are audibly transparent... 256 VBR or 320

Listen to the extra dirty 128kbps file I posted. It sounds like significantly more than regular 128kbps artifacts, there's some obnoxious distortion. I'm curious if this is the case with aac as well.
 
I'm fairly confident that the high bitrate aac files will not have a problem converting back and forth from wav, given that there wasn't a problem converting between them and high bitrate mp3, as well as high bitrate mp3 and wav.

 

[bigshot]

You may be sure of that, but prove it to the folks who have hard drives full of FLACs!

 

[chewy4]

Yeah it could be worth posting, it's just one extra file. I just don't want to overwhelm people with a million files covering every possibility, then none get listened to.

 

[chewy4]

Alright, so that crazy distortion does not exist when converting between 128kbps aac and wav. Even if I use the file that's been transcoded in between aac and mp3 a bunch of times already.
 
Here are the results from converting 320kbps aac to wav and back a bunch of times:
 

Spoiler: Spectograms

 

 
 

 
 
Here are the files(not volume matched):
 
 

Spoiler: Test%20files

 
Clean(320kbps aac converted once from wav): http://www.mediafire.com/?e4xlwl93cdtwfl9
Dirty(320kbps aac converted back and forth several times from wav): http://www.mediafire.com/?cdg4pbosd3f73xl
 

 

[bigshot]

Here is a question... do you think that perhaps the greatest change in the spectrum is when it's first compressed, and any further compression remains pretty much the same? Or do you think it's just a very tiny bit of loss each time and it would take a hundred reencodings to build up to a significant change?

 

[mikeaj]

Quote:

Here is a question... do you think that perhaps the greatest change in the spectrum is when it's first compressed, and any further compression remains pretty much the same? Or do you think it's just a very tiny bit of loss each time and it would take a hundred reencodings to build up to a significant change?

 
Main difference in the spectrum (which really doesn't allow you to visualize most significant differences in the sound, if any) is from the lowpass filtering, so yes the first change would be the greatest on the spectrum plot.
 
I'm not at all familiar with the psychoacoustic modeling and precise information storage of these codecs, but one would expect that the first transcoding essentially throws away the kind of information that subsequent transcodings would have thrown away.  Aside from some kind of noise or error signal accumulating, I don't think there should be too much different for passes after the first.
 
I think I'll look into this in some more detail, later.

 

[chewy4]

Quote:

Here is a question... do you think that perhaps the greatest change in the spectrum is when it's first compressed, and any further compression remains pretty much the same? Or do you think it's just a very tiny bit of loss each time and it would take a hundred reencodings to build up to a significant change?

There's a tiny bit of loss every time. This is definitely true as you can see it clearly in the spectograms. But it does take a lot for it to add up.
 
Also:
 
 
foo_abx 1.3.4 report
foobar2000 v1.1.14a
2013/02/27 23:00:22
 
File A: C:\Users\Computer\Desktop\blind testing\transcoding test\clean\VD_320AACNOTRANSCODING.m4a
File B: C:\Users\Computer\Desktop\blind testing\transcoding test\with wav\VD_320AACWITHWAV.m4a
 
23:00:22 : Test started.
23:00:47 : 01/01  50.0%
23:01:04 : 02/02  25.0%
23:01:25 : 03/03  12.5%
23:01:43 : 04/04  6.3%
23:02:08 : 05/05  3.1%
23:02:24 : 06/06  1.6%
23:02:57 : 07/07  0.8%
23:03:10 : 08/08  0.4%
23:04:05 : 09/09  0.2%
23:04:16 : 10/10  0.1%
23:04:26 : 11/11  0.0%
23:04:33 : Test finished.
 
 ---------- 
Total: 11/11 (0.0%)
 
 
I wouldn't consider this blatantly obvious, it just seemed like the treble parts were quieter(while the deep notes on the cello sounded the same). I did make sure to take my own advice and use replaygain on the test.
 
I still haven't done a really critical listen between the mp3>aac transcoding, I just looked for really obvious stuff before and didn't see any.