Very Serious WA5.3 Sound Quality Issue: Alias Frequencies (Professionally Measured)

kzuse · 29th September 2006, 14:57

I have just installed the new version 5.3 of Winamp. First I really have to say: GREAT WORK! Especially the lots of improvements (e.g. the speed improvements when recognizing dead links, and the overall speed in the library) are VERY GOOD!!!

But there is one - actually small, but VERY important - thing that will make me use the previous version again:

=== Positive part of this post ends here. ===

THE SOUND QUALITY OF MP3 PLAYBACK IS WORSE THAN IN ANY PREVIOUS VERSION!!! This ist the most important thing of an MP3 player I think, and it was one of the reason why I pasty loved Winamp: It just sounded better than the others (WMP & Co.).

But as with version 5.3, something must have changed in the in_mp3.dll. Winamp adds very annoying alias frequencies to all output. It has nothing to do whether I enable or disable the new "dithering" or "24 bit" options, or whether I use out_ds or out_wave.

In most pieces, you will not even hear it - okay. But in classical music or especially: in vocal music it can be heard very clearly.

That cannot be the future quality standard! Even WMP sounds better!

In the company in which I work (a german studio equipment company who built almost all european tv studios) we have software and hardware for performing spectral analysis.

For those who do not believe me the presence of the alias frequencies, I made them visible:

In the first picture (spectrum_original.png) you can see the spectral histogram of the song "Time" by the Alan Parsons Project (around the point of c. 02'45"). Played from an external MP3 player and recorded on the spectral analyzer's harddisc:

(Click image to enlarge)

You will definitely see the characteristic synthesizer line which makes a "pitch bend" upwards. It consists of several harmonics. But in the original, they ALL go up.

In the second picture (spectrum_winamp.png) you see the same song PLAYED BY WINAMP and recorded in the spectral analyzer:

(Click image to enlarge)

In the winamp playback, we have one other line above the others! And at its end, it is even going DOWN!!! This is a "mirror flipped" band of the frequency spectrum, copied into the signal from top->downwards. This is called "aliasing" and is a result of wrong sample calculations in the MP3 decoder.

What I want to say is: PLEASE offer a fix for in_mp3.dll very fast! (Question: Until then, can I use the old one from 5.24, or will winamp crash?)

It is certainly an in_mp3 error, WAV and WMA files don't show this issue!

Winamp must sound like before!! PLEASE!

best regards,
kzuse

Benski · 29th September 2006, 15:10

bug-fixed 64bit precision ASM optimized version
http://stashbox.org/322/in_mp3.dll

[edit] removed links to other 'potential fix' versions

kzuse · 29th September 2006, 15:47

Thanks! All of your files work and seem to solve the problem!!

Although I couldn't test it now using the equipment, but my ears and wavelab's analyzer tell me the issue has been fixed! A big thank you!

But which of the files would you recommend?

Regards
kzuse

Benski · 29th September 2006, 15:48

Quote:

Originally posted by kzuse
But which of the files would you recommend?

Bottom one is the fastest

kzuse · 29th September 2006, 15:54

Okay.. so I'll take that one (first I tended to take the third one, because it's smaller

).

Now, what was the error in the old 5.3 one (and what did you change in the new ones)?
And, what did originally cause it (more precise: What has been changed from 5.24 -> 5.3 in in_mp3.dll)?

Would be just interesting...

Benski · 29th September 2006, 15:57

Between 5.24 and 5.3 I rewrote the decoder to use 64bit floats internally, for better precision (2.7-5.24 had a noise floor of -86dB because of precision errors). I also made it switchable (via some clever #defines) between 32bit and 64bit precision, just to test quality differences.

Basically one line of optimized assembly wasn't set up properly to switch between 32bit and 64bit precision (it was using 32bit memory offsets always, even when compiled in 64bit precision mode). Unfortunately, I did most of the testing with the unoptimized C code, and had switched some parts back to assembly without as thorough of testing.

kzuse · 29th September 2006, 19:20

Thanks for the explanation. (And also thanks for the updated 1st post in the 5.3 released thread).

Because sound quality is a major thing in an MP3 player software, maybe the downloadeble installers should be recompiled to include the new in_mp3.dll? (of course without increasing the version number...) Just a suggestion..

Again thanks very much, now I can really enjoy Winamp again

Regards,
kzuse

Ojay · 29th September 2006, 20:09

Quote:

Originally posted by kzuse
Because sound quality is a major thing in an MP3 player software, maybe the downloadeble installers should be recompiled to include the new in_mp3.dll? (of course without increasing the version number...) Just a suggestion..

That is a temporary solution only for NEW downloaders. To get ALL WA 5.3 downloaders you would need to upload WA 5.31. But that will happen anyway soon with such a showstopper bug inside the most-used WA plugin...

DrO · 29th September 2006, 20:14

it's not really a showstopper of a bug now is it. if mp3 playback was completely broken then yes i'd have to agree. as for a 5.31 we shall have to wait and see...

-daz

armstrong · 8th October 2006, 14:36

Benski, thanks. Your quick fix is the best sounding in_mp3 from Nullsoft to date!

My personal lesson from that assembly incident is simple: tiny audio mistakes are hard to catch by ear, yet simple IMD tests (like the one kzuse ran) spot them right away. Makers of audio hardware knew this all along and test for IMD (and THD) routinely... there's hardly an excuse for software developers not to do the same.

Here's a good example: libmad (still my reference mp3 decoder for studio work) was tesed for IMD during development -- and sounds great. And a counterexample? the WinDoese kmixer... 'nuf said

Metalden · 10th October 2006, 05:17

I'm trying to download the fix for Sound quality issues...but the link won't open...Please help.

Sawg · 10th October 2006, 05:21

The site does not appear to be working right now. Just try again tomorrow.

Metalden · 10th October 2006, 05:36

Thank's for the help I'll try tomorrow...the krackiling sound is Killing Me.....

kzuse · 10th October 2006, 09:55

Please DO NOT USE the linked file in Benski's post above !!!

This is an older version, which does fix the sound quality issue, but still has some bugs regarding VBR files and gapless playback.

Please do ONLY use the file linked in this post:
http://forums.winamp.com/showthread....=256207#update

Or, use the direct link to the new dll file:
http://stashbox.org/419/in_mp3.dll

Have fun!
Best regards,

kzuse

29th September 2006, 14:57	#1
kzuse Senior Member Join Date: Oct 2005 Location: (D) Posts: 367	Very Serious WA5.3 Sound Quality Issue: Alias Frequencies (Professionally Measured) I have just installed the new version 5.3 of Winamp. First I really have to say: GREAT WORK! Especially the lots of improvements (e.g. the speed improvements when recognizing dead links, and the overall speed in the library) are VERY GOOD!!! But there is one - actually small, but VERY important - thing that will make me use the previous version again: === Positive part of this post ends here. === THE SOUND QUALITY OF MP3 PLAYBACK IS WORSE THAN IN ANY PREVIOUS VERSION!!! This ist the most important thing of an MP3 player I think, and it was one of the reason why I pasty loved Winamp: It just sounded better than the others (WMP & Co.). But as with version 5.3, something must have changed in the in_mp3.dll. Winamp adds very annoying alias frequencies to all output. It has nothing to do whether I enable or disable the new "dithering" or "24 bit" options, or whether I use out_ds or out_wave. In most pieces, you will not even hear it - okay. But in classical music or especially: in vocal music it can be heard very clearly. That cannot be the future quality standard! Even WMP sounds better! In the company in which I work (a german studio equipment company who built almost all european tv studios) we have software and hardware for performing spectral analysis. For those who do not believe me the presence of the alias frequencies, I made them visible: In the first picture (spectrum_original.png) you can see the spectral histogram of the song "Time" by the Alan Parsons Project (around the point of c. 02'45"). Played from an external MP3 player and recorded on the spectral analyzer's harddisc: (Click image to enlarge) You will definitely see the characteristic synthesizer line which makes a "pitch bend" upwards. It consists of several harmonics. But in the original, they ALL go up. In the second picture (spectrum_winamp.png) you see the same song PLAYED BY WINAMP and recorded in the spectral analyzer: (Click image to enlarge) In the winamp playback, we have one other line above the others! And at its end, it is even going DOWN!!! This is a "mirror flipped" band of the frequency spectrum, copied into the signal from top->downwards. This is called "aliasing" and is a result of wrong sample calculations in the MP3 decoder. What I want to say is: PLEASE offer a fix for in_mp3.dll very fast! (Question: Until then, can I use the old one from 5.24, or will winamp crash?) It is certainly an in_mp3 error, WAV and WMA files don't show this issue! Winamp must sound like before!! PLEASE! best regards, kzuse

8th October 2006, 14:36	#10
armstrong Junior Member Join Date: Oct 2006 Location: 32Â°4'1''N, 34Â°46'13''E Posts: 7	Testing audio software (and hardware) Benski, thanks. Your quick fix is the best sounding in_mp3 from Nullsoft to date! My personal lesson from that assembly incident is simple: tiny audio mistakes are hard to catch by ear, yet simple IMD tests (like the one kzuse ran) spot them right away. Makers of audio hardware knew this all along and test for IMD (and THD) routinely... there's hardly an excuse for software developers not to do the same. Here's a good example: libmad (still my reference mp3 decoder for studio work) was tesed for IMD during development -- and sounds great. And a counterexample? the WinDoese kmixer... 'nuf said

10th October 2006, 05:17	#11
Metalden Junior Member Join Date: Oct 2006 Posts: 3	I CAN'T OPEN THE FIX LINK I'm trying to download the fix for Sound quality issues...but the link won't open...Please help.

10th October 2006, 05:21	#12
Sawg Forum King Join Date: Jun 2000 Location: Phoenix, AZ Posts: 7,451	The site does not appear to be working right now. Just try again tomorrow. \| Brought to you by ^V ^C \|

Thread Tools	Search this Thread
Show Printable Version Email this Page	Search this Thread: Advanced Search
Display Modes
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode

29th September 2006, 15:10	#2
Benski Ben Allison Lead Developer Winamp Dev Team Join Date: Jan 2005 Location: Ashburn, VA Posts: 1,018	bug-fixed 64bit precision ASM optimized version http://stashbox.org/322/in_mp3.dll [edit] removed links to other 'potential fix' versions

29th September 2006, 15:47	#3
kzuse Senior Member Join Date: Oct 2005 Location: (D) Posts: 367	Thanks! All of your files work and seem to solve the problem!! Although I couldn't test it now using the equipment, but my ears and wavelab's analyzer tell me the issue has been fixed! A big thank you! But which of the files would you recommend? Regards kzuse

29th September 2006, 15:54	#5
kzuse Senior Member Join Date: Oct 2005 Location: (D) Posts: 367	Okay.. so I'll take that one (first I tended to take the third one, because it's smaller ). Now, what was the error in the old 5.3 one (and what did you change in the new ones)? And, what did originally cause it (more precise: What has been changed from 5.24 -> 5.3 in in_mp3.dll)? Would be just interesting...

29th September 2006, 15:57	#6
Benski Ben Allison Lead Developer Winamp Dev Team Join Date: Jan 2005 Location: Ashburn, VA Posts: 1,018	Between 5.24 and 5.3 I rewrote the decoder to use 64bit floats internally, for better precision (2.7-5.24 had a noise floor of -86dB because of precision errors). I also made it switchable (via some clever #defines) between 32bit and 64bit precision, just to test quality differences. Basically one line of optimized assembly wasn't set up properly to switch between 32bit and 64bit precision (it was using 32bit memory offsets always, even when compiled in 64bit precision mode). Unfortunately, I did most of the testing with the unoptimized C code, and had switched some parts back to assembly without as thorough of testing.

29th September 2006, 19:20	#7
kzuse Senior Member Join Date: Oct 2005 Location: (D) Posts: 367	Thanks for the explanation. (And also thanks for the updated 1st post in the 5.3 released thread). Because sound quality is a major thing in an MP3 player software, maybe the downloadeble installers should be recompiled to include the new in_mp3.dll? (of course without increasing the version number...) Just a suggestion.. Again thanks very much, now I can really enjoy Winamp again Regards, kzuse

29th September 2006, 20:14	#9
DrO - Join Date: Sep 2003 Location: UK Posts: 22,278	it's not really a showstopper of a bug now is it. if mp3 playback was completely broken then yes i'd have to agree. as for a 5.31 we shall have to wait and see... -daz

10th October 2006, 05:36	#13
Metalden Junior Member Join Date: Oct 2006 Posts: 3	Thank's for the help I'll try tomorrow...the krackiling sound is Killing Me.....

10th October 2006, 09:55	#14
kzuse Senior Member Join Date: Oct 2005 Location: (D) Posts: 367	Please DO NOT USE the linked file in Benski's post above !!! This is an older version, which does fix the sound quality issue, but still has some bugs regarding VBR files and gapless playback. Please do ONLY use the file linked in this post: http://forums.winamp.com/showthread....=256207#update Or, use the direct link to the new dll file: http://stashbox.org/419/in_mp3.dll Have fun! Best regards, kzuse