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: lossyWAV 1.1.0 Development Thread. (Read 182017 times) previous topic - next topic
0 Members and 2 Guests are viewing this topic.

lossyWAV 1.1.0 Development Thread.

Reply #50
I've been thinking about the difference between the replaygain / wavegain values calculated for the .lwcdf.wav file with --shaping 0 and --shaping 1 (spreadsheet attached).

There seems to be an average 12dB difference between the two (--shaping 1 = higher gain), although how this relates to bits to remove, I still need to determine.
lossyWAV -q X -a 4 -s h -A --feedback 2 --limit 15848 --scale 0.5 | FLAC -5 -e -p -b 512 -P=4096 -S- (having set foobar to output 24-bit PCM; scaling by 0.5 gives the ANS headroom to work)

lossyWAV 1.1.0 Development Thread.

Reply #51
I think that' s very natural and shows that your kind of noise shaping is effective.
The replaygain technique computes the perceived loudness. So replaygain says the perceived loudness of the error file is considerably lower when using noise shaping. So it confirms in an objective way what we can hear subjectively: noise shaping reduces the perceived noise - when listening to the noise.
Unfortunately it doesn't tell about masking, and so a noise shaped version isn't necessarily better when listening to the music.
With a high quality setting I feel pretty safe to use noise shaping, that's why I do it. The lower the quality setting however the more the risk IMO that noise concentrated in certain frequency regions can hurt (though it may well be that the benefits outweigh the risk also with lower quality settings).
lame3995o -Q1.7 --lowpass 17

lossyWAV 1.1.0 Development Thread.

Reply #52
There is a somthing up.
I'm hearing heavily distorted sound at -q0.
Wavpack -hh or TAK -pMax
OggVorbis aoTuVb6.03 -q 4

lossyWAV 1.1.0 Development Thread.

Reply #53
There is a somthing up.
I'm hearing heavily distorted sound at -q0.
Thanks for the sample, at what position in the audio are you hearing the distortion?

Also, what is the significance of the images?
lossyWAV -q X -a 4 -s h -A --feedback 2 --limit 15848 --scale 0.5 | FLAC -5 -e -p -b 512 -P=4096 -S- (having set foobar to output 24-bit PCM; scaling by 0.5 gives the ANS headroom to work)

lossyWAV 1.1.0 Development Thread.

Reply #54
I confirm,
I can hear it too sec 01, sec 07 & sec 12,
between sec 01 & sec 02 it is very very noisy in the back. I can ABX it 100%.

lossyWAV 1.1.0 Development Thread.

Reply #55
There is a somthing up.
I'm hearing heavily distorted sound at -q0.
Thanks for the sample, at what position in the audio are you hearing the distortion? ...

I just abxed sec. 6.4-8.6 8/10 (and I'm sure those who know this track will do better). I wouldn't call this spot heavily distorted though, but that's a matter of taste.
We know -q 0 isn't a real quality mode (though usually it's quite okay). My personal favorite for low bitrate is -q 1.5. It's not significantly higher in bitrate (312 kbps on average for my regular music test set) but to me quality does a little jump there. The spot I mentioned with your track, Mardel, is ok to me using -q 1.5.

Mardel and sauvage78, do you mind trying -q 1.5?
lame3995o -Q1.7 --lowpass 17

lossyWAV 1.1.0 Development Thread.

Reply #56
There is a somthing up.
I'm hearing heavily distorted sound at -q0.
Thanks for the sample, at what position in the audio are you hearing the distortion?

Also, what is the significance of the images?

I hear some sizzle in the high pitch range.
The essence of the picture, that 12 khz full is with sounds like that what they were not originally there.
Wavpack -hh or TAK -pMax
OggVorbis aoTuVb6.03 -q 4

lossyWAV 1.1.0 Development Thread.

Reply #57
... The essence of the picture, that 12 khz full is with sounds like that what they were not originally there.

Sure. lossyWAV does add noise. The essential question is: is it audible? We can't read that from a picture.
But with your sample we can hear it when using -q 0.
lame3995o -Q1.7 --lowpass 17

lossyWAV 1.1.0 Development Thread.

Reply #58
tried quickly v1.0.1f -q 1.5, the artefact is much reduced but still easyly ABXable IMHO
I wouldn't have thought only a 1.5 increase would have reduced it so much, so I think it should go away on higher setting ... futher test is required.

I was using -q 0 too, all I can say is that I don't want to use -q 0 anymore, for me it was as easy to ABX as MP3 64Kbps on castanets sample, a nightmare

Typo: I typed -q 1 intead of -q 0 several times, it's fixed now

lossyWAV 1.1.0 Development Thread.

Reply #59
... The essence of the picture, that 12 khz full is with sounds like that what they were not originally there.
Sure. lossyWAV does add noise. The essential question is: is it audible? We can't read that from a picture.
But with your sample we can hear it when using -q 0.
Maybe now is the time to revisit the -spf string (again) and maybe tighten up its response in the 12kHz to 16kHz region. I'll get to work re-introducing the -spf parameter (as --spf <string>) and post v1.0.1g this evening.

Thanks for the sample and listening input sauvage78 and Mardel (and, of course halb27).
lossyWAV -q X -a 4 -s h -A --feedback 2 --limit 15848 --scale 0.5 | FLAC -5 -e -p -b 512 -P=4096 -S- (having set foobar to output 24-bit PCM; scaling by 0.5 gives the ANS headroom to work)

 

lossyWAV 1.1.0 Development Thread.

Reply #60
tried quickly v1.0.1f -q 1.5, the artefact is much reduced but still easyly ABXable IMHO ...

Thanks for your test. Glad to hear that -q 1.5 is better for you though still easily ABXable. Your ears are better than mine.
We should keep in mind that general transparency is expected at -q 5, at least for people with good hearing. Due to a series of internal precautions it's not necessarily -q 5 which is to use, and we can stay a bit lower.

It would be interesting to learn at what quality level this sample becomes transparent to everybody who tries.


Maybe now is the time to revisit the -spf string (again) and maybe tighten up its response in the 12kHz to 16kHz region. ....

Please keep in mind that all the tests so far have been performed with very low quality settings.
When introducing a more restrictive -spf string for the highest frequency zone (you did that already with your latest spf change) bitrate will go up, and we must compare the results to the results of the current spf setting at a correspondingly higher quality setting.

Perhaps it would be wise to find out first what current quality setting is necessary for generally accepted transparency of this sample.
In case transparency can be achieved with a rather low current quality setting I think we shouldn't mind and just call it a problem sample we should be willing to accept with very low quality settings.
In case transparency is achieved not before ~ -q 5 I think it's worth investigating whether a change in the machinery can bring a progress.
lame3995o -Q1.7 --lowpass 17

lossyWAV 1.1.0 Development Thread.

Reply #61
I agree with halb27, I have spend the last 20 min doing serious ABX & it happened that -q1.5 was much harder to ABX than what I first thought ... I can ABX -q 0 100% without serious listening, I was forced to listen carefully  to ABX -q 1.5, it was placebo when I said it was easy ... I was finally able to ABX with near 100% success but it was not easy at all in the end ... I can ABX -q 0 with only the 2 first seconds ... I was forced to concentrate & listen up to second 7 to ABX -q1.5 ... & in the end it was very little added noise to an already noisy original sample for -q 1.5 while it's LOTS of noise for -q 0 ... so I was not self-confident even if I finally succeeded.

I think it is very specific to -q 0 & already very noisy source. -q 0 is not safe anymore for me but everything is not wasted. I am 100% sure that this sample is transparent for me far before -q 5 now as I doubt I can ABX even -q 2 ... but I won't test I am bored now

lossyWAV 1.1.0 Development Thread.

Reply #62
I agree with halb27, I have spend the last 20 min doing serious ABX & it happened that -q1.5 was much harder to ABX than what I first thought ... I can ABX -q 0 100% without serious listening, I was forced to listen carefully  to ABX -q 1.5, it was placebo when I said it was easy ... I was finally able to ABX with near 100% success but it was not easy at all in the end ... I can ABX -q 0 with only the 2 first seconds ... I was forced to concentrate & listen up to second 7 to ABX -q1.5 ... & in the end it was very little added noise to an already noisy original sample for -q 1 while it's LOTS of noise for -q 0 ... so I was not self-confident even if I finally succeeded.

I think it is very specific to -q 0 & already very noisy source. -q 0 is not safe anymore for me but everything is not wasted.

Glad to hear this. Do you mind going up a bit in quality level to learn when this track is transparent for you?
lame3995o -Q1.7 --lowpass 17

lossyWAV 1.1.0 Development Thread.

Reply #63
Sorry, I just edited that I wouldn't test -q 2 now ... maybe I will try tomorow cause it's bedtime for me now.

Don't expect miracle, I am french but I am no Guruboolez

lossyWAV 1.1.0 Development Thread.

Reply #64
Sorry, I just edited that I wouldn't test -q 2 now ... maybe I will try tomorow cause it's bedtime for me now.

Don't expect miracle, I am french but I am no Guruboolez

Sure, but your testing is very welcome. Thank you, and have a good night.
lame3995o -Q1.7 --lowpass 17

lossyWAV 1.1.0 Development Thread.

Reply #65
Sorry, I just edited that I wouldn't test -q 2 now ... maybe I will try tomorow cause it's bedtime for me now.

Don't expect miracle, I am french but I am no Guruboolez
Thanks for all of the input - I am very interested in your -q 2 ABX results.

Nick.
lossyWAV -q X -a 4 -s h -A --feedback 2 --limit 15848 --scale 0.5 | FLAC -5 -e -p -b 512 -P=4096 -S- (having set foobar to output 24-bit PCM; scaling by 0.5 gives the ANS headroom to work)

lossyWAV 1.1.0 Development Thread.

Reply #66
I've been thinking about -q 0 and 1 ("my" quality levels as my ears are a bit shot....). As I have been refining the reference_threshold > threshold_index > bits_to_remove calculations, I had a few kbit/s to play with which I have recycled into the quality of the output of the lower quality presets.

I have re-implemented the --lowpass parameter to allow user with better ears to raise the upper limit of the range of frequencies lossyWAV uses in its analyses.

I have tweaked the --spf and minimum_bits_to_keep:
Code: [Select]
  spreading_function_string         : string[precalc_analyses*(spread_zones+2)-1]='22222-22222-22223-12224-12234-12345';
  quality_minimum_bits_to_keep      : array[0..Quality_Presets] of double = (3.000,3.000,3.000,3.125,3.250,3.375,3.500,3.625,3.750,3.875,4.000);
and get the following:
Code: [Select]
|-----------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|
| lossyWAV  | FLAC  | -q 10 | -q 9  | -q 8  | -q 7  | -q 6  | -q 5  | -q 4  | -q 3  | -q 2  | -q 1  | -q 0  |
|-----------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|
|1.0.1b     |784kbps|637kbps|607kbps|577kbps|545kbps|513kbps|480kbps|449kbps|427kbps|390kbps|349kbps|306kbps|
|-----------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|
|1.0.1g     |784kbps|639kbps|610kbps|580kbps|548kbps|516kbps|484kbps|453kbps|430kbps|396kbps|354kbps|320kbps|
|-----------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|
|1.0.1g --LC|784kbps|656kbps|628kbps|600kbps|569kbps|538kbps|506kbps|474kbps|450kbps|414kbps|369kbps|334kbps|
|-----------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|

lossyWAV beta 1.0.1g attached to post #1 in this thread.
lossyWAV -q X -a 4 -s h -A --feedback 2 --limit 15848 --scale 0.5 | FLAC -5 -e -p -b 512 -P=4096 -S- (having set foobar to output 24-bit PCM; scaling by 0.5 gives the ANS headroom to work)

lossyWAV 1.1.0 Development Thread.

Reply #67
This morning I tested the problem sample at V1.0.1f -q 0/-q 0.5/-q 1/-q 1.5 with my speaker (Logitech Z3-I) instead of my headphone (Philips SHP1900) it was easier to ABX but I dunno if that is because I was tired yesterday or because of the cheap headphone (Mainboard is Asus A7N8X-Deluxe) this time I have keep my log to backup my claims as I done it very seriously.

foo_abx 1.3.3 report
foobar2000 v0.9.5.2
2008/05/22 09:56:09

File A: C:\Documents and Settings\Mes documents\02- Test\Nouveau dossier\b.lossless.flac
File B: C:\Documents and Settings\Mes documents\02- Test\Nouveau dossier\b.lossyQ0.0.flac

09:56:09 : Test started.
09:56:37 : 01/01  50.0%
09:56:50 : 02/02  25.0%
09:57:02 : 03/03  12.5%
09:57:18 : 04/04  6.3%
09:57:31 : 05/05  3.1%
09:57:45 : 06/06  1.6%
09:57:58 : 07/07  0.8%
09:58:09 : 08/08  0.4%
09:58:20 : 09/09  0.2%
09:58:33 : 10/10  0.1%
09:58:47 : 11/11  0.0%
09:58:58 : 12/12  0.0%
09:59:30 : Test finished.

----------
Total: 12/12 (0.0%)

foo_abx 1.3.3 report
foobar2000 v0.9.5.2
2008/05/22 10:01:16

File A: C:\Documents and Settings\Mes documents\02- Test\Nouveau dossier\b.lossless.flac
File B: C:\Documents and Settings\Mes documents\02- Test\Nouveau dossier\b.lossyQ0.5.flac

10:01:16 : Test started.
10:01:38 : 01/01  50.0%
10:01:53 : 02/02  25.0%
10:02:06 : 03/03  12.5%
10:02:19 : 04/04  6.3%
10:02:38 : 05/05  3.1%
10:02:51 : 06/06  1.6%
10:03:50 : 07/07  0.8%
10:04:01 : 08/08  0.4%
10:04:23 : 09/09  0.2%
10:04:36 : 10/10  0.1%
10:04:48 : 11/11  0.0%
10:04:59 : 12/12  0.0%
10:05:02 : Test finished.

----------
Total: 12/12 (0.0%)

foo_abx 1.3.3 report
foobar2000 v0.9.5.2
2008/05/22 10:06:22

File A: C:\Documents and Settings\Mes documents\02- Test\Nouveau dossier\b.lossless.flac
File B: C:\Documents and Settings\Mes documents\02- Test\Nouveau dossier\b.lossyQ1.0.flac

10:06:22 : Test started.
10:06:50 : 01/01  50.0%
10:07:20 : 02/02  25.0%
10:07:36 : 03/03  12.5%
10:07:47 : 04/04  6.3%
10:08:00 : 05/05  3.1%
10:08:24 : 06/06  1.6%
10:08:45 : 07/07  0.8%
10:09:06 : 08/08  0.4%
10:09:21 : 09/09  0.2%
10:09:52 : 10/10  0.1%
10:10:17 : 11/11  0.0%
10:10:36 : 12/12  0.0%
10:10:46 : Test finished.

----------
Total: 12/12 (0.0%)

foo_abx 1.3.3 report
foobar2000 v0.9.5.2
2008/05/22 10:20:41

File A: C:\Documents and Settings\Mes documents\02- Test\Nouveau dossier\b.lossless.flac
File B: C:\Documents and Settings\Mes documents\02- Test\Nouveau dossier\b.lossyQ1.5.flac

10:20:41 : Test started.
10:21:01 : 01/01  50.0%
10:21:24 : 02/02  25.0%
10:21:54 : 03/03  12.5%
10:22:34 : 04/04  6.3%
10:24:09 : 05/05  3.1%
10:25:48 : 06/06  1.6%
10:26:20 : 07/07  0.8%
10:27:01 : 08/08  0.4%
10:27:21 : 09/09  0.2%
10:28:15 : 10/10  0.1%
10:29:33 : 11/11  0.0%
10:30:18 : 12/12  0.0%
10:30:22 : Test finished.

----------
Total: 12/12 (0.0%)

Will test -q 2 later as it took me 10min just for -q 1.5 but it becomes hard. -q 0 & -q 0.5 were easy (no hesitation), -q 1 medium (a few hesitations) & -q 1.5 hard (several hesitations), I was forced to re-listen several times at -q 1.5 but unlike yesterday I was able to ABX all with only the first 2 second & was not forced to listen up to second 7.

lossyWAV 1.1.0 Development Thread.

Reply #68
I can ABX -q 2 too but it takes 23min with a pause in the middle ... I will not try to ABX further it's too hard.

foo_abx 1.3.3 report
foobar2000 v0.9.5.2
2008/05/22 11:08:32

File A: C:\Documents and Settings\Mes documents\02- Test\b.lossless.wav
File B: C:\Documents and Settings\Mes documents\02- Test\b.lossyQ2.0.wav

11:08:32 : Test started.
11:10:25 : 01/01  50.0%
11:12:44 : 02/02  25.0%
11:13:49 : 03/03  12.5%
11:14:53 : 04/04  6.3%
11:16:12 : 05/05  3.1%
11:19:18 : 06/06  1.6%
11:22:21 : 07/07  0.8%
11:23:00 : 08/08  0.4%
11:23:57 : 09/09  0.2%
11:24:57 : 10/10  0.1%
11:28:47 : 11/11  0.0%
11:31:24 : 12/12  0.0%
11:31:28 : Test finished.

----------
Total: 12/12 (0.0%)

In the end I would describe it like this:
-q 0.0 & -q0.5 added noise/distortion, clear artefact/flaw
-q 1.5 & -q2.0 tearing in the original distortion, I don't ABX an added artefact, I ABX either the lenght of the distortion or the depht of the tearing (it's slightly longer/deeper)
-q1.0 is in the middle: slightly added noise, slighly longer distortion.

lossyWAV 1.1.0 Development Thread.

Reply #69
I have re-implemented the --lowpass parameter to allow user with better ears to raise the upper limit of the range of frequencies lossyWAV uses in its analyses.
THX. -q 2.5 --lowpass 17000 is a very good sound quality for my sample. I tried q 1; q 1.5;  q 2; q 2 lowpass 16k, 17k, 18, 19, 20k; q 2.5; q2.5 lowpass16k and 17k. I haven't heard any sound artifacts at -q 2.5 --lowpass 17000.
Wavpack -hh or TAK -pMax
OggVorbis aoTuVb6.03 -q 4

lossyWAV 1.1.0 Development Thread.

Reply #70
I can ABX -q 2 too but it takes 23min with a pause in the middle ... I will not try to ABX further it's too hard.

In the end I would describe it like this:
-q 0.0 & -q0.5 added noise/distortion, clear artefact/flaw
-q 1.5 & -q2.0 tearing in the original distortion, I don't ABX an added artefact, I ABX either the lenght of the distortion or the depht of the tearing (it's slightly longer/deeper)
-q1.0 is in the middle: slightly added noise, slighly longer distortion.
Many thanks for your efforts. As you can see above, I have modified the HF end of the spreading function in 1.0.1g. Also, you might be interested in using the --lowpass parameter to increase the upper limit of the range of frequencies lossyWAV uses in its analyses.

When your ears have recovered, I would appreciate it if you could listen to 1.0.1g -q 0 (it probably will still not require ABX, however I would be interested to hear if there is any improvement). Maybe slowly increasing --lowpass from 16000 to, say, 18000 will improve this particular sample (at the expense of FLAC bitrate).

Thanks again,

Nick.
lossyWAV -q X -a 4 -s h -A --feedback 2 --limit 15848 --scale 0.5 | FLAC -5 -e -p -b 512 -P=4096 -S- (having set foobar to output 24-bit PCM; scaling by 0.5 gives the ANS headroom to work)

lossyWAV 1.1.0 Development Thread.

Reply #71
Would it be rude to request a reality check here? No lossy codec should be transparent at the very bottom of its quality scale.

I think it's very useful to have at least one non-transparent setting at the bottom of the range - it lets people hear what kind of artefacts lossyWAV is adding. It also confirm to the average user that the quality scale works, and stops people from saying "even q0 is transparent - what happens if you lower the bitrate still further".

Cheers,
David.

lossyWAV 1.1.0 Development Thread.

Reply #72
Code: [Select]
-s, --scale <n>  scaling factor from WaveGain, etc; default=1.000000; n<>0!

LossyWav doesn't permit n to be a positive scale factor either, but many of my classical and easy listening piano albums require it.
My question: can it be changed in a future version ? My Wavegain.exe can do it but that's another run in the process.

lossyWAV 1.1.0 Development Thread.

Reply #73
Would it be rude to request a reality check here? No lossy codec should be transparent at the very bottom of its quality scale.

I think it's very useful to have at least one non-transparent setting at the bottom of the range - it lets people hear what kind of artefacts lossyWAV is adding. It also confirm to the average user that the quality scale works, and stops people from saying "even q0 is transparent - what happens if you lower the bitrate still further".

Cheers,
David.
Not at all - it's a requirement of development.... I agree with what you're saying, though maybe there is some scope for improving the performance at -q 0 a bit....

Code: [Select]
-s, --scale <n>  scaling factor from WaveGain, etc; default=1.000000; n<>0!

LossyWav doesn't permit n to be a positive scale factor either, but many of my classical and easy listening piano albums require it.
My question: can it be changed in a future version ? My Wavegain.exe can do it but that's another run in the process.
At present you would have to wavegain the file before processing with lossyWAV. What upper limit do you think would be reasonable, as it's an easy fix (1 value changed in the nparameters unit)?
lossyWAV -q X -a 4 -s h -A --feedback 2 --limit 15848 --scale 0.5 | FLAC -5 -e -p -b 512 -P=4096 -S- (having set foobar to output 24-bit PCM; scaling by 0.5 gives the ANS headroom to work)

lossyWAV 1.1.0 Development Thread.

Reply #74
If you wavegain at more than unity, you can introduce clipping, but more importantly the efficiency could be reduced (probably not by much in practice).