You have permission to ravage me sexually at your leisure.

Wow. When using this in foobar, when I set "Highest BPS Mode Supported" to 32, I get a 58kbps output file. When I set it to 16bits, I get a 75kbps output file! What gives?

[edit] This is true before the patch as well. q0. Source was a 161kbps vorbis file from b5.61 (32bit).

Noise added due to dither or truncating 32->16 bits.

Is there any way to avoid this noise during transcoding?  Other than using 32bits?  Is there some foobar setting?

Rarewares changelog cites 'Multichannel code improved'. A while ago Vorbis was rather poor at multichannel encoding (compared to AAC, for example). Have I missed some major change in that direction?

Hi

I encounter 2 problems, first one:

There is a lot of clipping in vorbis from -q5 to -q10. You can see it on screenshots (in all screenshots first wave is source wav, second vorbis -q8 encoded with aoTuV 5.61).
Encoding to vorbis files that are very close to clipping gives output file with actual clipping, is there maybe a way to tell encoder that if wave goes comes close to clipping that it is possible to allocate more bits in order to avoid it?. I know you can use ReplayGain to avoid clipping, but it`s not default option, i`v made package of source and test files from -q5 to -q8 (aoTuV 5.61), the *.ogg files have calculated peak value in tags and you can simply play them in foobar2000, put volume a bit up and see how much of clipping is inside that files, for comparison You can use source file or apply ReplayGain - avoid clipping according to peak by track.


Test Files

Hi

I encounter 2 problems, first one:

there is a lot of clipping in vorbis from -q5 to -q10, You can see it on screenshots (in all screenshots first wave is source wav, second vorbis -q8 encoded with aoTuV 5.61).
Encoding to vorbis files that are very close to clipping gives output file with actual clipping, is there maby a way to tell encoder that if wave goes to close to clipping, have higher bitrate to avoid it.
I know You can use ReplayGain to avoid clipping, but it`s not default option, i`v made package of source and test files from -q5 to -q8 (aoTuV 5.61), the *.ogg files have calculated peak value in tags and You can simply play them in foobar2000, put volume a bit up and see how much of clipping is inside that files, for comparison You can use source file or aply ReplayGain - avoid clipping accordin to peak by track.

foo_abx 1.3.1 report
foobar2000 v0.9.5.1
2009/01/08 23:22:53

File A: D:\clipping\udial.wav
File B: D:\clipping\udial-q5.ogg

23:22:53 : Test started.
23:24:06 : 01/01  50.0%
23:25:16 : 02/02  25.0%
23:25:34 : 03/03  12.5%
23:25:43 : 04/04  6.3%
23:25:53 : 05/05  3.1%
23:26:04 : 06/06  1.6%
23:26:23 : 07/07  0.8%
23:26:59 : 08/08  0.4%
23:27:04 : Test finished.

foo_abx 1.3.1 report
foobar2000 v0.9.5.1
2009/01/08 23:27:30

File A: D:\clipping\udial.wav
File B: D:\clipping\udial-q6.ogg

23:27:30 : Test started.
23:27:49 : 01/01  50.0%
23:28:02 : 02/02  25.0%
23:28:14 : 03/03  12.5%
23:28:25 : 04/04  6.3%
23:29:31 : 05/05  3.1%
23:29:54 : 06/06  1.6%
23:30:03 : 07/07  0.8%
23:30:15 : 08/08  0.4%
23:30:18 : Test finished.

 ----------
Total: 8/8 (0.4%)

foo_abx 1.3.1 report
foobar2000 v0.9.5.1
2009/01/08 23:31:54

File A: D:\clipping\udial.wav
File B: D:\clipping\udial-q7.ogg

23:31:54 : Test started.
23:32:19 : 01/01  50.0%
23:32:25 : 02/02  25.0%
23:32:34 : 03/03  12.5%
23:32:50 : 04/04  6.3%
23:32:56 : 05/05  3.1%
23:33:06 : 06/06  1.6%
23:33:13 : 07/07  0.8%
23:33:23 : 08/08  0.4%
23:33:24 : Test finished.

 ----------
Total: 8/8 (0.4%)

foo_abx 1.3.1 report
foobar2000 v0.9.5.1
2009/01/08 23:34:46

File A: D:\clipping\udial.wav
File B: D:\clipping\udial-q8.ogg

23:34:46 : Test started.
23:34:58 : 01/01  50.0%
23:35:07 : 02/02  25.0%
23:36:33 : 03/03  12.5%
23:36:39 : 04/04  6.3%
23:36:46 : 05/05  3.1%
23:36:54 : 06/06  1.6%
23:36:59 : 07/07  0.8%
23:37:08 : 08/08  0.4%
23:37:11 : Test finished.

 ----------
Total: 8/8 (0.4%)

foo_abx 1.3.1 report
foobar2000 v0.9.5.1
2009/01/08 23:38:01

File A: D:\clipping\udial.wav
File B: D:\clipping\udial-q9.ogg

23:38:01 : Test started.
23:38:23 : 01/01  50.0%
23:38:29 : 02/02  25.0%
23:38:44 : 03/03  12.5%
23:38:52 : 04/04  6.3%
23:39:00 : 05/05  3.1%
23:39:09 : 06/06  1.6%
23:39:19 : 07/07  0.8%
23:39:24 : 08/08  0.4%
23:39:26 : Test finished.

 ----------
Total: 8/8 (0.4%)

foo_abx 1.3.1 report
foobar2000 v0.9.5.1
2009/01/08 23:40:20

File A: D:\clipping\udial.wav
File B: D:\clipping\udial-q10.ogg

23:40:20 : Test started.
23:41:25 : 01/01  50.0%
23:41:32 : 02/02  25.0%
23:41:44 : 03/03  12.5%
23:41:54 : 04/04  6.3%
23:42:08 : 05/05  3.1%
23:42:21 : 06/06  1.6%
23:42:30 : 07/07  0.8%
23:42:37 : 08/08  0.4%
23:42:40 : Test finished.

 ----------
Total: 8/8 (0.4%)

I am not going to rule out the possibility that the encoder could be causing clipping, but I find it highly unlikely this is a real serious problem. You can't "magically" throw more bits at the encoder to have it prevent clipping. The test samples aren't transient in nature just high frequency tones. You can only allocate more bits to deal with artifacts like pre-echo or microattacks. The one artifact that people have been complaining about recently is "distortion". Whether or not it's related to this is anyone's guess. I am guessing to adjust that you would have to alter some code in psy.c, but Aoyumi probably know's more about the internals then I do. Maybe the noise floor was altered slighly during the tuning process? your sample was clipped before hand so we don't even know if it's a real problem.

Does all this mean that aoTuV encoded ogg vorbis has higher volume than the source?

All of this was done with replaygain off.
With replaygain on in the abx tester, I don't think I can ABX it, but there's a strange noise (sounds kinda like an alien ray gun in some cheesy low-budget sci-fi film) on both the lossless and lossy copies (WINE bug? foobar abx plugin and diskwrriter plugin applying track gain while completely disregarding the maximum gain that can be applied without causing clipping?).

Quote from: ameyer17 on 2009-01-09 06:15:13

All of this was done with replaygain off.
With replaygain on in the abx tester, I don't think I can ABX it, but there's a strange noise (sounds kinda like an alien ray gun in some cheesy low-budget sci-fi film) on both the lossless and lossy copies (WINE bug? foobar abx plugin and diskwrriter plugin applying track gain while completely disregarding the maximum gain that can be applied without causing clipping?).

No, there is no wine bug, in that smple there is 19-21 kHz sin wave very close to clipping, that frequency on that volume makes some hardware to generate distortions, only thing I can say is congrats on Your hearing
For me that frequency in that sample is audiable at pre-amp volume 0,0 dB and master volume -1,2 dB.

I encounter 2 problems, first one:

there is a lot of clipping in vorbis from -q5 to -q10, You can see it on screenshots (in all screenshots first wave is source wav, second vorbis -q8 encoded with aoTuV 5.61).
Encoding to vorbis files that are very close to clipping gives output file with actual clipping, is there maby a way to tell encoder that if wave goes to close to clipping, have higher bitrate to avoid it.
I know You can use ReplayGain to avoid clipping, but it`s not default option, i`v made package of source and test files from -q5 to -q8 (aoTuV 5.61), the *.ogg files have calculated peak value in tags and You can simply play them in foobar2000, put volume a bit up and see how much of clipping is inside that files, for comparison You can use source file or aply ReplayGain - avoid clipping accordin to peak by track.

of course that is only what I think is happening and it`s my guess why encoder adds clipping.

Is this the first sample ever that's ABXable at -q 10 but not at -q 3?

...for me using low pass filter is out of the question...

Quote from: dev on 2009-01-12 10:29:59

...for me using low pass filter is out of the question...

Why? Can you provide ABX test results confirming that you can hear the difference between no lowpass and, say, a 16KHz lowpass in typical music content? Why make the encoder's job harder than it needs to be if you can't hear any difference... or can you?

Cheers, Slipstreem. 

No, I can`t but my head splits in two when I see cutoff`s on spectrum analizer, let say it`s a matter of taste

And about low pass filter; it works in this sample couse the clipping accoures in high frequency if there were clipping on vocals, guitar, etc..., than it would do any good.

Quote from: dev on 2009-01-12 17:05:54

No, I can`t but my head splits in two when I see cutoff`s on spectrum analizer, let say it`s a matter of taste

So, in audible terms, it means nothing to you. Graphs also mean nothing unless you listen with your eyes.

Quote

And about low pass filter; it works in this sample couse the clipping accoures in high frequency if there were clipping on vocals, guitar, etc..., than it would do any good.

So you don't think it makes sense to do something that has no audible effect to you personally in terms of HF content but helps the encoder in some problem cases? Interesting.

Cheers, Slipstreem.