Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: recommended using VBR model for CBR? (Read 29091 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

recommended using VBR model for CBR?

Reply #25 –
Quote from: memomai on
Exactly what I'm looking for: is there a better CBR mode than -b XXX ?...Example: has "-V0 -b 256 -B 256 (and following paramters for forcing CBR header and so on)" got better quality than -b 256? where 256 is just an example for the high bitrate range.

Thanks for clarification.
Quote from: memomai on
I'd like to leave out the question of compatibility first (= does "-V0 -b256 -B256" give me the same overall compatibility, e.g. streaming, like -b 256?)

???? If this is not your concern, then why narrow focus only to CBR?

I can't really tell about audio quality for the exact question within your questioning framework. My target was finding a perfect or at least better quality than when using -V0 for certain problem samples when going higher in average bitrate compared to -V0. My listening results for this target were negative when using V0 -b xxx -B yyy. But your question is at least a little bit different.

As for the irrelevant non-audio data I just encoded my test set of pop tracks of various kind and used Omion's great mp3packer afterwards to squeeze out the garbage and get an impression of the real audio content:

Result for -b 320: average bitrate is 304 kbps.
Result for -V0 -b 320: average bitrate is 233 kbps. Audio data amount is so much lower than when using -b 320 that I personally would not be motivated to give this setting a try.
Result for -b 256: average bitrate is 253 kbps.
Result for -V0 -b 256 -B 256: average bitrate is 229 kbps. The discrepancy in audio data amount compared to using -b 256 is not so extreme as in the case of CBR 320. But it's still remarkable.
lame3995o -Q1.7 --lowpass 17
 

recommended using VBR model for CBR?

Reply #26 –
I have done a track that has a major precho problem at the start, which produces a noise that sounds like a water sprinkler or rain. I have tested it with V0, V0 -b320 and -b 320.

V0

Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6
2008/12/09 21:59:48

File A: C:\Rips\Skinny Puppy - VIVIsectVI\04. Human Disease (S.K.U.M.M).flac
File B: C:\Temp\Human Disease (S.K.U.M.M) V0.mp3

21:59:48 : Test started.
22:00:29 : 01/01  50.0%
22:00:38 : 02/02  25.0%
22:00:50 : 03/03  12.5%
22:00:55 : 04/04  6.3%
22:01:00 : 05/05  3.1%
22:01:27 : 06/06  1.6%
22:01:45 : 07/07  0.8%
22:01:53 : 08/08  0.4%
22:02:16 : 09/09  0.2%
22:02:31 : 10/10  0.1%
22:02:33 : Test finished.

 ---------- 
Total: 10/10 (0.1%)

V0 -b 320

Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6
2008/12/09 22:05:55

File A: C:\Rips\Skinny Puppy - VIVIsectVI\04. Human Disease (S.K.U.M.M).flac
File B: C:\Temp\Human Disease (S.K.U.M.M) V0 -b 320.mp3

22:05:55 : Test started.
22:06:46 : 01/01  50.0%
22:06:56 : 02/02  25.0%
22:07:26 : 03/03  12.5%
22:07:36 : 04/04  6.3%
22:07:50 : 05/05  3.1%
22:08:04 : 06/06  1.6%
22:08:28 : 07/07  0.8%
22:08:41 : 08/08  0.4%
22:08:53 : 09/09  0.2%
22:09:09 : 10/10  0.1%
22:09:11 : Test finished.

 ---------- 
Total: 10/10 (0.1%)

The bitrate is at 320, which makes me wonder if -b 320 is for min frames or a bug on LAME 3.98, but am not sure. For some reason at the time when i did this test, it was alot more easy to ABX then plain V0. I tried to ABX both v0 and v0 -b 320, but i had no luck passing atm since i am getting tired.

-b 320

Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6
2008/12/09 22:27:20

File A: C:\Rips\Skinny Puppy - VIVIsectVI\04. Human Disease (S.K.U.M.M).flac
File B: C:\Temp\Human Disease (S.K.U.M.M) -b 320.mp3

22:27:20 : Test started.
22:28:24 : 01/01  50.0%
22:28:34 : 02/02  25.0%
22:29:06 : 02/03  50.0%
22:29:17 : 03/04  31.3%
22:29:27 : 04/05  18.8%
22:29:39 : 04/06  34.4%
22:30:07 : 05/07  22.7%
22:30:16 : 06/08  14.5%
22:30:36 : 06/09  25.4%
22:30:44 : 07/10  17.2%
22:30:48 : 08/11  11.3%
22:31:07 : 09/12  7.3%
22:31:26 : 10/13  4.6%
22:31:35 : 10/14  9.0%
22:31:46 : 11/15  5.9%
22:31:51 : 11/16  10.5%
22:32:03 : 11/17  16.6%
22:32:09 : 11/18  24.0%
22:32:16 : 12/19  18.0%
22:32:19 : 13/20  13.2%
22:32:28 : 14/21  9.5%
22:32:49 : 15/22  6.7%
22:33:02 : 16/23  4.7%
22:33:12 : 17/24  3.2%
22:33:28 : 18/25  2.2%
22:33:38 : 18/26  3.8%
22:33:48 : 19/27  2.6%
22:34:03 : 20/28  1.8%
22:34:15 : 21/29  1.2%
22:34:21 : 22/30  0.8%
22:34:49 : 23/31  0.5%
22:35:04 : 24/32  0.4%
22:35:11 : 25/33  0.2%
22:35:37 : Test finished.

 ---------- 
Total: 25/33 (0.2%)

Better then V0 and V0 -b 320 IMO, very hard to ABX since i had to really concentrate with this one.
"I never thought I'd see this much candy in one mission!"

recommended using VBR model for CBR?

Reply #27 –
I have uploaded a 30 sec sample of that track
"I never thought I'd see this much candy in one mission!"

recommended using VBR model for CBR?

Reply #28 –
Quote from: /mnt on
I have done a track that has a major precho problem at the start, which produces a noise that sounds like a water sprinkler or rain. I have tested it with V0, V0 -b320 and -b 320. ... (-b 320) Better then V0 and V0 -b 320 IMO, very hard to ABX since i had to really concentrate with this one.

Thanks for your test.
Out of curiosity: do you mind trying --abr 285 --lowpass 16.5 on your sample? I'm interested how it compares against -b 320 and -V0.
lame3995o -Q1.7 --lowpass 17

recommended using VBR model for CBR?

Reply #29 –
I've taken my favourite sample lately and encoded it at 224kbps because Lame 3.98.3 -V2 produces 222kbps output and is easy to ABX. I encoded with these settings:
  • lame -b 224
  • lame -V 0 --vbr-new -b 224 -B 224
Both encodings produced 224 kbps output.

First FLAC against CBR 224:
Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6
2008/12/09 19:00:52

File A: F:\Temp\01 - Pet Shop Boys - In The Night.flac
File B: F:\Temp\01 - Pet Shop Boys - In The Night --398CBR224.mp3

19:00:52 : Test started.
19:01:31 : 01/01  50.0%
19:01:43 : 02/02  25.0%
19:01:49 : 03/03  12.5%
19:01:53 : 04/04  6.3%
19:01:56 : 05/05  3.1%
19:02:22 : 06/06  1.6%
19:02:27 : 07/07  0.8%
19:02:36 : 08/08  0.4%
19:02:46 : 09/09  0.2%
19:03:06 : 10/10  0.1%
19:03:59 : 11/11  0.0%
19:04:04 : 12/12  0.0%
19:04:21 : 13/13  0.0%
19:04:32 : 14/14  0.0%
19:04:39 : 15/15  0.0%
19:04:45 : Test finished.

 ---------- 
Total: 15/15 (0.0%)


Then ABX'ed -V 0 with minimum and maximum bitrates set at 224 kbps:
Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6
2008/12/10 09:44:14

File A: F:\Temp\01 - Pet Shop Boys - In The Night.flac
File B: F:\Temp\01 - Pet Shop Boys - In The Night --398V0b224B224.mp3

09:44:14 : Test started.
09:44:49 : 01/01  50.0%
09:45:07 : 02/02  25.0%
09:45:22 : 03/03  12.5%
09:45:30 : 03/04  31.3%
09:45:37 : 04/05  18.8%
09:45:53 : 05/06  10.9%
09:46:01 : 05/07  22.7%
09:46:11 : 05/08  36.3%
09:46:33 : 06/09  25.4%
09:47:33 : 06/10  37.7%
09:47:45 : 06/11  50.0%
09:48:50 : 06/12  61.3%
09:49:23 : Test finished.

 ---------- 
Total: 6/12 (61.3%)

I've tried and noticed something on high frequencies of snares but I don't seem to be able to ABX (posted result is from third and last attempt).

Then I ABX'ed both lossy files:
Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6
2008/12/10 10:18:44

File A: F:\Temp\01 - Pet Shop Boys - In The Night --398CBR224.mp3
File B: F:\Temp\01 - Pet Shop Boys - In The Night --398V0b224B224.mp3

10:18:44 : Test started.
10:19:27 : 01/01  50.0%
10:19:37 : 02/02  25.0%
10:19:44 : 03/03  12.5%
10:19:50 : 04/04  6.3%
10:20:02 : 05/05  3.1%
10:20:17 : 06/06  1.6%
10:20:59 : 07/07  0.8%
10:21:37 : 08/08  0.4%
10:21:49 : 09/09  0.2%
10:22:02 : 10/10  0.1%
10:22:22 : 11/11  0.0%
10:22:30 : Test finished.

 ---------- 
Total: 11/11 (0.0%)

Being a bit annoyed by these results and the fact that I can ABX this sample also with standard -V 0 (278kbps) I tried again ABX'ing FLAC against -V 0 -b 224 -B 224 and after getting 1/5 I quitted (got tired listening).

My conclusion based on this sample is that -V 0 -b 224 -B 224 encoding is much closer to lossless than CBR 224 is, actually it seems transparant 

Edit: Yesterday I've also tried to ABX flac against lame3.98 CBR 320kbps to see if at maximum mp3 bitrate the distortions are still there. At 320kbps the problematic spots seems to have become transparent.

recommended using VBR model for CBR?

Reply #30 –
Quote from: Alexxander on
... I can ABX this sample also with standard -V 0 (278kbps) I tried again ABX'ing FLAC against -V 0 -b 224 -B 224 and after getting 1/5 I quitted (got tired listening). ...

Things are getting real interesting.
lame3995o -Q1.7 --lowpass 17

recommended using VBR model for CBR?

Reply #31 –
Quote
???? If this is not your concern, then why narrow focus only to CBR?


halb27: I wrote FIRST, I mean when I know more about using VBR model in CBR, then I check this point of compatibilty. I just want to know the answer to the question of quality, then I'll go to the question of compatibility.

Hope this clarified my statement.
FB2K,APE&LAME

recommended using VBR model for CBR?

Reply #32 –
Quote from: Alexxander on
... and the fact that I can ABX this sample also with standard -V 0 (278kbps)...

I'm pretty sure I've posted the abx result somewhere but can't find it so here goes the one I did in 4 months ago:
Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.5.5
2008/08/18 16:45:56

File A: F:\Temp\01 - Pet Shop Boys - In The Night.flac
File B: F:\Temp\01 - Pet Shop Boys - In The Night --398V0.mp3

16:45:56 : Test started.
16:46:03 : 01/01  50.0%
16:46:11 : 02/02  25.0%
16:46:17 : 03/03  12.5%
16:46:23 : 04/04  6.3%
16:46:33 : 05/05  3.1%
16:46:42 : 06/06  1.6%
16:46:58 : 07/07  0.8%
16:47:07 : Test finished.

 ---------- 
Total: 7/7 (0.8%)

recommended using VBR model for CBR?

Reply #33 –
Quote from: memomai on
... I just want to know the answer to the question of quality, then I'll go to the question of compatibility. ...

Compatibility shouldn't be an issue. It is when using just -b xxx -B xxx as 32 kbps frames are used for silence, but there is a switch forcing that this isn't done. Haven't tried though.


Quote from: Alexxander on
...I'm pretty sure I've posted the abx result somewhere but can't find it so here goes the one I did in 4 months ago:...

Oh, you didn't test -V0 and -V0 -b224 -B224 in close temporary neighborhood. This relativates things a bit (I know from my own listening tests that sensitivity towards problems can vary).
Do you mind comparing -V0 and -V0 -b 224 -B 224 directly? I'm a bit upset if -V0 were worse, so I'd welcome if you could back up your statement by comparing at the same time.
lame3995o -Q1.7 --lowpass 17

recommended using VBR model for CBR?

Reply #34 –
vbr for cbr isn't so trivial. forcing cbr like constraints on vbr will not make quality much better. e.g. V9 -b320 is still using V9 PSY Parameters + different code level tunings and NS to vanilla -b320

recommended using VBR model for CBR?

Reply #35 –
Quote from: shadowking on
vbr for cbr isn't so trivial. forcing cbr like constraints on vbr will not make quality much better. e.g. V9 -b320 is still using V9 PSY Parameters + different code level tunings and NS to vanilla -b320


I've already said that.

Quote
Do you mind comparing -V0 and -V0 -b 224 -B 224 directly? I'm a bit upset if -V0 were worse, so I'd welcome if you could back up your statement by comparing at the same time.


I am interested in that too. Up to now, I was quite sure that forcing a bitrate wouldn't produce a significative change. Now you say that a variant of the same setting (-V 0) produces both, less bitrate and higher quality.

This is even more strange provided that you forced a high bitrate limit. This has always been discouraged, since the VBR algorithm wasn't (isn't?) using the bit reservoir in the same way that CBR does, so CBR 256 could use more bits for a specific frame than -V0 -b256 -B 256.

/mnt thanks for your results. Those are more in line of what halb27 and I were saying.

recommended using VBR model for CBR?

Reply #36 –
Quote from: halb27 on
...Do you mind comparing -V0 and -V0 -b 224 -B 224 directly? I'm a bit upset if -V0 were worse, so I'd welcome if you could back up your statement by comparing at the same time.

I just tried it very hard and no luck: 12/19 and 6/12. This made me a bit angry so I tried -V0 against FLAC and no luck neither: 7/13 and 6/15. I guess this needs the best of me, I will try again in the next few days.

recommended using VBR model for CBR?

Reply #37 –
Thanks a lot.
I know what you're feeling. I had it way too often that I heard (subtle) problems with real regular music with just careful listening and could ABX them, and other times (the majority of times) when I couldn't.
lame3995o -Q1.7 --lowpass 17

recommended using VBR model for CBR?

Reply #38 –
Yesterday I got a bit angry because after failing to ABX -V0 against -V0 -b 224 -B 224 I tried to ABX -V0 against FLAC and failed while earlier this year I didn't. Today my hearing hasn't dissapointed me:

Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6
2008/12/11 12:30:16

File A: F:\Temp\01 - Pet Shop Boys - In The Night.flac
File B: F:\Temp\01 - Pet Shop Boys - In The Night --398V0.mp3

12:30:16 : Test started.
12:30:31 : 01/01  50.0%
12:30:42 : 02/02  25.0%
12:30:51 : 03/03  12.5%
12:31:24 : 04/04  6.3%
12:32:05 : 05/05  3.1%
12:32:24 : 06/06  1.6%
12:33:14 : 06/07  6.3%
12:33:33 : 06/08  14.5%
12:33:41 : 07/09  9.0%
12:33:46 : 08/10  5.5%
12:33:49 : 08/11  11.3%
12:33:55 : 08/12  19.4%
12:34:02 : 09/13  13.3%
12:34:20 : 10/14  9.0%
12:34:28 : 10/15  15.1%
12:34:31 : 11/16  10.5%
12:34:34 : 12/17  7.2%
12:34:42 : 13/18  4.8%
12:34:52 : 14/19  3.2%
12:35:02 : 15/20  2.1%
12:35:09 : 16/21  1.3%
12:35:14 : 17/22  0.8%
12:35:22 : 18/23  0.5%
12:35:34 : 19/24  0.3%
12:35:46 : 20/25  0.2%
12:36:03 : 21/26  0.1%
12:36:24 : 21/27  0.3%
12:36:59 : 22/28  0.2%
12:37:11 : 23/29  0.1%
12:37:23 : 24/30  0.1%
12:37:42 : 25/31  0.0%
12:37:50 : Test finished.

 ---------- 
Total: 25/31 (0.0%)


and standard -V0 against -V0 -b 224 -B 224:
Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6
2008/12/11 16:26:03

File A: F:\Temp\01 - Pet Shop Boys - In The Night --398V0.mp3
File B: F:\Temp\01 - Pet Shop Boys - In The Night --398V0b224B224.mp3

16:26:03 : Test started.
16:26:41 : 01/01  50.0%
16:26:51 : 02/02  25.0%
16:27:08 : 03/03  12.5%
16:27:25 : 04/04  6.3%
16:27:56 : 05/05  3.1%
16:28:11 : 05/06  10.9%
16:28:29 : 05/07  22.7%
16:28:51 : 06/08  14.5%
16:29:05 : 07/09  9.0%
16:29:22 : 07/10  17.2%
16:29:33 : 08/11  11.3%
16:29:43 : 09/12  7.3%
16:30:18 : 10/13  4.6%
16:30:32 : 11/14  2.9%
16:31:14 : 11/15  5.9%
16:31:40 : 12/16  3.8%
16:31:56 : 13/17  2.5%
16:32:13 : 13/18  4.8%
16:32:31 : 14/19  3.2%
16:32:43 : 15/20  2.1%
16:33:03 : 16/21  1.3%
16:33:48 : 16/22  2.6%
16:34:01 : 16/23  4.7%
16:34:21 : 16/24  7.6%
16:34:38 : 17/25  5.4%
16:34:53 : 18/26  3.8%
16:35:02 : 19/27  2.6%
16:36:20 : 19/28  4.4%
16:36:37 : 20/29  3.1%
16:36:43 : 21/30  2.1%
16:37:00 : 22/31  1.5%
16:37:26 : 23/32  1.0%
16:37:48 : 24/33  0.7%
16:37:59 : Test finished.

 ---------- 
Total: 24/33 (0.7%)

This test was very tough and exhausting to ABX because both sound almost the same. Of the 3 problem spots I've used (all 3 between 2.0 and 11.5 secs after track start) in this ABX test I could only use the one at around 10.3 sec.

So this makes me wonder why with a much lower bitrate I get similar quality/tranparency. If -V 0 setting is more easy to ABX against FLAC than using -V 0 with a higher than standard minimum bitrate and lower than standard maximum bitrate I would conclude that the distortion in -V 0 is caused by using a too low minimum bitrate (and not by not having bitrates above 320kbps).

I know this is only one sample and a personal conclusion. What do you think?

recommended using VBR model for CBR?

Reply #39 –
Quote from: Alexxander on
...So this makes me wonder why with a much lower bitrate I get similar quality/tranparency. If -V 0 setting is more easy to ABX against FLAC than using -V 0 with a higher than standard minimum bitrate and lower than standard maximum bitrate I would conclude that the distortion in -V 0 is caused by using a too low minimum bitrate (and not by not having bitrates above 320kbps).

I know this is only one sample and a personal conclusion. What do you think?

Thanks a lot for your hard test.
By just looking from the outside at your ABX results it looks like both encodings have roughly the same quality. You got at for instance 14/19 with both candidates. You were faster doing so with -V0, so ABXing -V0 was easier.

Peace of mind comes back to me as I would really be upset if -V0 wasn't transparent whereas -V0 -b 224 -B 224 was.
You mentioned bitrate being significantly lower with -b 224 -B 224. Can you tell a bit about it?

Anyway using a CBR 224 setting the VBR way seems to be attractive as you showed already that it outperforms -b 224 which was the OP's question.

Guess I'll play around with it this evening.
My main concern with -B 224 is about behavior when the V0 machinery needs a certain amount of data for the encoded audiocontent of a frame and doesn't get the space for it due to the -B restriction. Is this situation handled well? Such a behavior can be forced by using something like -V0 -B 128. If the VBR machinery works well also with the frame size restriction, quality should be adequate for 128 kbps, comparable to CBR 128. Will try tonight.
lame3995o -Q1.7 --lowpass 17

recommended using VBR model for CBR?

Reply #40 –
Quote from: halb27 on
Quote from: /mnt on

I have done a track that has a major precho problem at the start, which produces a noise that sounds like a water sprinkler or rain. I have tested it with V0, V0 -b320 and -b 320. ... (-b 320) Better then V0 and V0 -b 320 IMO, very hard to ABX since i had to really concentrate with this one.

Thanks for your test.
Out of curiosity: do you mind trying --abr 285 --lowpass 16.5 on your sample? I'm interested how it compares against -b 320 and -V0.


Sorry I didn't have time to read the whole discussion, I just want to share some of my findings.

(I can't provide listening tests, these are just my subjective findings which might not be right)

The CBR psymodel is very good at pre-echo. I have actually found the new VBR psymodel producing pretty bad preecho artifacts, I posted abx tests in some ol thread. So I can understand that -V0 -b320 is inferior here to -b320.

In my "Lame CBR seems to be broken" post, I concluded the quality is pretty bad when using 128kbs CBR and lower. Lame has problems with tonal distortion, added "crumpling" noise and dropouts.
However, I didn't say anything about very high bitrate CBR. It might very well be the case that the mentioned artifacts disappear with high bitrate cbr.

recommended using VBR model for CBR?

Reply #41 –
Here are some ABX tests i've been playing around.

The samples are some of those that I linked to in this post:
http://www.hydrogenaudio.org/forums/index....st&p=590228

ABX -V0 -b 128 -B 128 against -V6 -b 128 -B 128:
Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6
2008/12/11 19:23:13

File A: D:\bigtest\v0b128B128 - 1 brainwash9-part1.mp3
File B: D:\bigtest\v6b128B128 - 1 brainwash9-part1.mp3

19:23:13 : Test started.
19:23:44 : 01/01  50.0%
19:23:47 : 02/02  25.0%
19:23:49 : 03/03  12.5%
19:23:52 : 04/04  6.3%
19:23:55 : 05/05  3.1%
19:23:58 : 06/06  1.6%
19:24:01 : 07/07  0.8%
19:24:03 : 08/08  0.4%
19:24:06 : 09/09  0.2%
19:24:09 : 10/10  0.1%
19:24:11 : 11/11  0.0%
19:24:14 : 12/12  0.0%
19:24:18 : 13/13  0.0%
19:24:20 : 14/14  0.0%
19:24:24 : 15/15  0.0%
19:24:29 : 16/16  0.0%
19:24:36 : Test finished.

 ---------- 
Total: 16/16 (0.0%)
(ABX range: 3.0-4.4)
This one was easy. and the -V0 sounds worse than the V6 encode. Concretely, this shows the same problem that low bitrate CBR is having:  Trying to keep too much quality with not enough bitrate. The -V 0 has dropouts and warbling, while the -V6 is not annoying.

ABX -V0 -b 224 -B 224 against -V6 -b 224 -B 224:
Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6
2008/12/11 19:26:12

File A: D:\bigtest\v0b224B224 - 1 brainwash9-part1.mp3
File B: D:\bigtest\v6b224B224 - 1 brainwash9-part1.mp3

19:26:12 : Test started.
19:27:23 : 01/01  50.0%
19:27:26 : 02/02  25.0%
19:27:28 : 03/03  12.5%
19:27:31 : 04/04  6.3%
19:27:34 : 05/05  3.1%
19:27:36 : 06/06  1.6%
19:27:39 : 07/07  0.8%
19:27:41 : 08/08  0.4%
19:27:43 : 09/09  0.2%
19:27:46 : 10/10  0.1%
19:27:48 : 11/11  0.0%
19:27:52 : 12/12  0.0%
19:27:55 : 13/13  0.0%
19:27:57 : 14/14  0.0%
19:28:00 : 15/15  0.0%
19:28:03 : 16/16  0.0%
19:28:05 : Test finished.

 ---------- 
Total: 16/16 (0.0%)
As one would expect, the -V6 encode was of lower quality than the -V0 encode. The -V 0 was trasparent (didn't try to ABX), while the -V 6 had dropouts/underwater.


ABX  -V6 -b 128 -B 128 against -V6 -b 224 -B 224:
Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6
2008/12/11 19:31:52

File A: D:\bigtest\v6b128B128 - 2 brainwash9-part2.mp3
File B: D:\bigtest\v6b224B224 - 2 brainwash9-part2.mp3

19:31:52 : Test started.
19:38:19 : 01/01  50.0%
19:38:21 : 02/02  25.0%
19:38:25 : 03/03  12.5%
19:38:30 : 04/04  6.3%
19:38:33 : 05/05  3.1%
19:38:43 : 06/06  1.6%
19:38:52 : 07/07  0.8%
19:39:01 : 08/08  0.4%
19:39:10 : 09/09  0.2%
19:39:20 : 10/10  0.1%
19:39:24 : 11/11  0.0%
19:39:28 : 12/12  0.0%
19:39:35 : 13/13  0.0%
19:39:39 : 14/14  0.0%
19:39:49 : 15/15  0.0%
19:39:52 : 16/16  0.0%
19:39:54 : Test finished.

 ---------- 
Total: 16/16 (0.0%)

I could detect only an improvement in preecho here. Yet, lame says it uses the same amount of long and short blocks.

I could not ABX this setting with brainwash9-part1. Here, the higher bitrate didn't help -V6 to have more quality.

I could also ABX gameplayv5-part1. In this one, there is less warbling/underwater with the higher bitrate.

ABX -V0 vs -V -b 224 -B 224
Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6
2008/12/11 20:07:01

File A: D:\bigtest\v0b224B224 - 5 gameplayv5-part2.mp3
File B: D:\bigtest\v0 - 5 gameplayv5-part2.mp3

20:07:01 : Test started.
20:08:06 : 01/01  50.0%
20:08:10 : 02/02  25.0%
20:08:26 : 03/03  12.5%
20:08:31 : 04/04  6.3%
20:08:34 : 05/05  3.1%
20:08:38 : 06/06  1.6%
20:08:41 : 07/07  0.8%
20:08:47 : 08/08  0.4%
20:08:50 : 09/09  0.2%
20:08:54 : 10/10  0.1%
20:08:57 : 11/11  0.0%
20:09:03 : 12/12  0.0%
20:09:08 : 13/13  0.0%
20:09:11 : 14/14  0.0%
20:09:21 : 15/15  0.0%
20:09:45 : 16/16  0.0%
20:09:47 : Test finished.

 ---------- 
Total: 16/16 (0.0%)
(ABXed 1.4-1.9)
-V0 is better with this sample than -V0 -b 224 -B 224.  -V0 bitrate is 278kbps.  I ABXed warbling/undewater in the range mentioned.


ABX -V0 -b 224 -B 224 vs CBR 224
Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6
2008/12/11 20:15:38

File A: D:\bigtest\v0b224B224 - 5 gameplayv5-part2.mp3
File B: D:\bigtest\b224 - 5 gameplayv5-part2.mp3

20:15:38 : Test started.
20:16:18 : 01/01  50.0%
20:16:21 : 02/02  25.0%
20:16:25 : 03/03  12.5%
20:16:28 : 04/04  6.3%
20:16:32 : 05/05  3.1%
20:16:34 : 06/06  1.6%
20:16:38 : 07/07  0.8%
20:16:52 : 08/08  0.4%
20:16:56 : 09/09  0.2%
20:17:04 : 10/10  0.1%
20:17:11 : 11/11  0.0%
20:17:15 : 12/12  0.0%
20:17:18 : 13/13  0.0%
20:17:22 : 14/14  0.0%
20:17:28 : 15/15  0.0%
20:17:32 : 16/16  0.0%
20:17:33 : Test finished.

 ---------- 
Total: 16/16 (0.0%)
(ABXed 1.4-1.9)
Since i had localized this problem range on the -V0 -b224 -B224, I wanted to compare it with the CBR encoding. The conclusion? The CBR encoding does not have this artifact.


Overall (limited to these tests) conclusions

Targeting a higher than default bitrate (via the -b and -B settings) on a -V setting can improve its quality. Even though can does not mean it will.

Targeting a lower than default bitrate (via the -b and -B settings) on a -V setting is almost guaranteed to cause lower quality than a lower -V setting targeted at that bitrate.

CBR xxx can sound better than -V 0 -b xxx -B xxx. (In this case, xxx = 224)

[Edit: changed code with codebox]

recommended using VBR model for CBR?

Reply #43 –
a) I tried -V0 -b 224 -B 224 with my typical track set and got at an average bitrate of 218 kbps after mp3packing. That is the air in the encoded file is negligible with this way of doing CBR 224.
Using -V0 -b 224 -B 224 -Y (there were members here saying CBR/ABR's HF behavior corresponds to the -Y switch) I get at 213 kbps after mp3packing. For a comparison: plain -V0 yields 239 kbps, -V0 -Y 218 kbps, the difference of which is significantly higher than the difference of 213 and 218 kbps. This looks favorably for the audio precision of -V0 -b 224 -B 224 -Y for the frequency range provided by -Y. Backs up a bit the beleive that -b 224 may contribute to this.
While this doen't tell about quality it gives an attractive and motivating framework for testing.

b) As for my concerns about what the encoder does when there is not enough space for encoding due to the -B restriction I did a short listening test on Birds, Castanets, harp40_1, herding_calls, and trumpet at 112 kbps which is much too low. As -b 112 has a 15.6 kHz lowpass I compared -b 112 against -V0 -b 112 -B 112 --lowpass 15.6.

Birds, harp40_1, and herding_calls were better with the VBR way of doing CBR 112, harp40_1 was even a lot better.
Castanets and trumpet were better with -b 112, trumpet a lot.

May be trumpet shows that the -B restriction can be very restrictive. Of course this is due to the fact that the storage space generally required by -V0 doesn't leave room for special needs. -V0 simply isn't adequate here.

Using say -V5 should be more adequate, that is leave more room for special situations even with the -B restriction. Improvement on trumpet however is negligible. It takes -V7 to get at a clearly noticeable improvement on trumpet (still a lot inferior to -b 112). On the other hand -V7 puts already a lot of air into the encoding as shown with mp3packer.

Because of the small improvement on trumpet when going from -V0 to -V7 I don't think the -B restriction is the decisive factor. I think it's just the differences between the VBR and the CBR machinery which gives different quality to the two methods depending on track.

Testing Birds, Castanets, harp40_1, herding_calls, and trumpet using -V0 -b 224 -B 224 -Y against -b 224:
It's the same story as with CBR 112, just at a good quality level: with castanets and trumpet -b 224 is better.
The other samples are alright to me with either mode.
Using -V0 -Y castanets is on par to me compared to using -b 224 as well as trumpet. I could not reliably ABX the difference of the -V0 -Y result against that of -b 224 (though my feelings are a tiny bit more with -V0 -Y for castanets and with -b 224 for trumpet).

So for my ears and these tracks I wouldn't prefer -V0 -b 224 -B 224 -Y over -b 224.
lame3995o -Q1.7 --lowpass 17