PDA

View Full Version : AC3Filter bug (AC3 decoding quality of liba52 much worse than Azid)

S_O
25th June 2003, 23:54
Edit: !!!! This is wrong, bug is in AC3Filter, not in liba52 !!!!


I found out both decoder have dramatically quality difference, this is very noticeable on low bitrates.
Here is one ac3-file (32kHz, 80kBit/s, Stereo):
maria.ac3 (290 KB) (http://l.b.oltmanns.bei.t-online.de/maria.ac3)
Decode this file first to wav using HeadAC3he/BeSweet (a program using Azid) and listen to the wav. It wonīt sound good, but now play the ac3 in Windows Media Player using AC3Filter (or any other program using liba52), here it sound completly differnt and to my ears much worse. I think liba52 has to be fixed.
tiki4
26th June 2003, 17:48
What do you expect, AFAIK liba52 is OpenSource.

[NOT that I say that open source has worse quality, no, but maybe it's just not as major as azid].

tiki4
S_O
26th June 2003, 21:48
Iīm sorry, the bug is _not_ in liba52, itīs in AC3Filter. I compiled a52dec (sample application of liba52) myself and I found out, here it sounds just perfect, the same like in RealPlayer (the file was originaly encoded with RealAudioEncoder) and Azid.

I donīt know why it sounds so ugly with AC3Filter, but this bug should be fixed very soon.
tiki4
27th June 2003, 10:16
Ha, now you got me. I should have checked that myself before stating that liba52 has worse quality than azid. Anyhow, I tested AC3filter quite some time but wasn't very happy with the quality and went back to the Moonlight Odio Dekoda (mlcom.ax) which never gave me any trouble.

Maybe we should contact Valex about that problem. Do you think you can send him an email where the problem is located?

tiki4
S_O
27th June 2003, 14:41
Maybe we should contact Valex about that problem. Do you think you can send him an email where the problem is located?
I already did it direrctly after finding the bug. This is a very serious bug and should be fixed very soon, otherwise it could damage the image of AC3Filter dramatically.
tiki4
27th June 2003, 15:02
Thanks :)

tiki4
rjamorim
9th July 2003, 21:37
Any news if the bug has been fixed with latest version? (0.69b)

I'm using OdioDekoda for the time being.
S_O
10th July 2003, 11:41
No, quality is still the same and Valex hasnīt answered my mail yet.
Valex
11th July 2003, 00:29
Hi!

@S_O
I'm just bask from a long expedition and just read your email. Thanks for your sample, I'll explore this problem near time.

PS. Please, next time use xvalex@mail.ru. @hotbox.ru is my backup address and I do not check it frequently....

@tiki
How exactly did you make your tests? Can you please, express what exactly didn't you like, and upload some samples with bad moments...
S_O
11th July 2003, 01:37
I yust analysed the spectral graph of a azid decode and one with ac3filter (using GraphEdit). They look quite different, in azid the low-pass seems to be 20kHz, but in ac3filter 17kHz. Several things in the decode of ac3filter seems to be missing (like this line in the picture Iīve posted), but also at lower frequences as you can see.
With both DRC disabled, azid decode is exactly double as load as ac3filter. I mixed the ac3filter-decode inverted at 200% in the azid-decode: Only artifacts are left. Not noise, artifacts:
http://l.b.oltmanns.bei.t-online.de/ac3filterbug.png
I donīt know from what this artifacts are, but that are definetly not rounding differences.
The source file is from DVD (Star Trek 6 trailer), 192kbps Stereo. If you want I can upload the file (2,99MB), but I think itīs the same with other ac3-files.
Valex
14th July 2003, 12:20
@S_O
Is it better now?
http://ac3filter.sourceforge.net/download/ac3filter_0_70b_rc1.exe
S_O
14th July 2003, 13:23
It is already much better now, but there is still a noticeable difference. I cannot hear any difference between RealPlayer and a52dec (azid is unabled to decode low-bitrate ac3), but ac3filter still sounds differnt (and worse). What did you change?

Edit: The test-file is 11025Hz; 16kbps; mono
Valex
18th August 2003, 11:43
Fixed a big and old bug in decoder! Try this one:

http://ac3filter.sourceforge.net/download/ac3filter_0_70b_rc2.exe
S_O
18th August 2003, 16:55
Great! Sounds perfect now!
tiki4
19th August 2003, 13:09
Sorry for not sharing my results in here for a while.

I'm going to test the new 0.7 version now and will post some results soon.

tiki4
bleo
20th August 2003, 02:29
um... what is this bug exactly? Does it only apply to stereo/low bitrate files? I've been using AC3Filter 0.69b to transcode 5.1 384/448kbps AC3s for my rips and would hate to have to do them all again!
Valex
20th August 2003, 04:00
@bleo
It is applied for low bitrates mainly. If some frequency coeffitients are not transmitted in the bitstream random noise is generated for that frequencies according to the rough PSD function transmitted. This noise was too weak for high frequencies (coupling channel to be exact). So some frequencies was almost totally out. It was leading to noticable 'metal sound' on low bitrates. For high bitrates problem exists but less noticable because almost all coeffitients are transmitted.

Problem exists on all 0.5x and 0.6x versions. So only judge if you need to recompress your movies is your ears. For example I cannot notice this difference on usual films...
Sirber
10th November 2003, 20:05
link down