lossyWAV 1.2.0 released, Added noise WAV bitdepth reduction method Options

[Nick.C]

Point taken. I have modified the source (I have never really stopped modifying it, if truth be told.... Ok, so I have a "habit" ) to initialise libfftw3f-3.dll first and if found to carry out all FFT calculations using data type Single rather than Double for libfftw3-3.dll. If libfftw3f-3.dll cannot be initialised then libfftw3-3.dll will be used if it exists - if neither can be initialised then processing falls back to internal FFT routines which are Double based. This will be released at some point - I am working on a couple of other additions as well. Thanks for the prod.

[Agent69]

Nick,

In your opinion, which lossless codec works best in conjunction with LossyWav? I know you use it with FLAC, but I don't know if that's because you prefer it or if that's simply a case of using what the Sanza supports.

[Nick.C]

I carried out a test using my 10 album test set processed at --portable --altpreset:

• FLAC: 385kbps;
• TAK: 361kbps;
• WavPack: 403kbps.

So, TAK wins the compression battle. I use FLAC, as you surmised, because of hardware compatibility.

[lvqcl]

LossyWAV 1.2.0 --standard:

WavPack -x1: 477.4 kbps
FLAC -5: 450.7 kbps
TAK -p2: 433.0 kbps

[Agent69]

Thanks a lot guys. I really appreciate your responses.

I am going to either use Flac or Wavpack. If I go with WavPack, I will probably use -x3, as using -h or -hh increases decoding effort on the part of the computer (if I understand things correctly).

This post has been edited by Agent69: Jan 21 2010, 04:20

[Steve Forte Rio]

Does anyone can to rewrite the batch file for using with TAKC? (lossyTAK.bat)
I've tried but commandline encoder can not find the temp wav files (lossy and lwcdf)

This post has been edited by Steve Forte Rio: Jan 23 2010, 11:16

[2E7AH]

here is example:
lossyTAK.bat

CODE

goto %1
goto end
:correction
c:\"Program Files"\encoders\lossywav %2 %4 %5 %6 %7 %8 %9 --%1 --below --silent
c:\"Program Files"\encoders\takc -e -p2m -fsl512 -silent -overwrite -lp "%~n2.lwcdf.wav" "%~n3.tak.lwcdf.tak"
c:\"Program Files"\encoders\takc -e -p2m -fsl512 -silent -overwrite -lp "%~n2.lossy.wav" "%~n3.tak" && del %~n2.lwcdf.wav %~n2.lossy.wav
goto end
:merge
if not exist "%~n3.lwcdf.tak" goto end
copy "%~n3" "%~n3.lossy.tak" /y
c:\"Program Files"\encoders\takc -d -fim2 -silent -overwrite "%~n3.lwcdf.tak" && c:\"Program Files"\encoders\takc -d -fim2 -silent -overwrite "%~n3.lossy.tak"
c:\"Program Files"\encoders\lossywav "%~n3.lossy.wav" -M --below --silent && del "%~n3.lossy.wav" "%~n3.lwcdf.wav" "%~n3.lossy.tak"
c:\"Program Files"\encoders\takc -e -p2m -silent -lp "%~n3.wav" "%~n3.tak" && del "%~n3.wav"
:end

correction part produces %filename%.lossy.tak and %filename%.lossy.tak.lwcdf.tak:


merge part produces %filename%.lossy.tak.tak file which is identical to original file:

All tracks decoded fine, no differences found.

[shadowking]

I've been playing with the new lossywav and also wavpack lossy. Lossywav peforms really good on very critical samples but i noticed on other 'easier' stuff its really easy to find problems on Q1. I decided to try Q2 tonight . I went for 2x5 trials.

Q2 (sorry i didn't keep log):

5/5 then 4/5 = 9/10 - added hiss on vocal

Q2.5 --portable

foo_abx 1.3.4 report
foobar2000 v0.9.6.8
2010/01/27 00:28:13

File A: C:\windows\profiles\My Documents\temp\Ivri Lider\The New People\09 - Dear Sir.flac
File B: C:\windows\profiles\My Documents\temp\09 - Dear Sir.lossy.flac

00:28:13 : Test started.
00:28:32 : 01/01 50.0%
00:28:39 : 02/02 25.0%
00:28:57 : 03/03 12.5%
00:29:24 : 04/04 6.3%
00:29:36 : 05/05 3.1%
00:33:09 : 06/06 1.6%
00:33:19 : 06/07 6.3%
00:34:16 : 06/08 14.5%
00:35:29 : 07/09 9.0%
00:36:14 : 08/10 5.5%
00:37:47 : Test finished.

----------
Total: 8/10 (5.5%)


The 1st 5 trials I was convinced and in full concentration . An added hiss on vocals no different than Q2. The 2nd half i was truck with fatigue.
I might have another go - is it significant ? 8/10 ?

Overall Q2 + 2.5 = 17/20

This post has been edited by shadowking: Jan 26 2010, 14:53

[shadowking]

i decided to try wavpack @ 250 and 300k


-b250x4:

foo_abx 1.3.4 report
foobar2000 v0.9.6.8
2010/01/27 00:59:46

File A: C:\windows\profiles\My Documents\temp\Ivri Lider\The New People\09 - Dear Sir.flac
File B: C:\windows\profiles\My Documents\temp\09 - Dear Sir.wv

00:59:46 : Test started.
01:00:07 : 01/01 50.0%
01:00:16 : 02/02 25.0%
01:00:32 : 02/03 50.0%
01:00:41 : 03/04 31.3%
01:01:09 : 04/05 18.8%
01:01:16 : 05/06 10.9%
01:01:21 : 06/07 6.3%
01:01:26 : 07/08 3.5%
01:01:44 : 08/09 2.0%
01:02:23 : 09/10 1.1%
01:02:26 : Test finished.

----------
Total: 9/10 (1.1%)



-b300x4:

foo_abx 1.3.4 report
foobar2000 v0.9.6.8
2010/01/27 01:04:22

File A: C:\windows\profiles\My Documents\temp\Ivri Lider\The New People\09 - Dear Sir.flac
File B: C:\windows\profiles\My Documents\temp\09 - Dear Sir.wv

01:04:22 : Test started.
01:04:38 : 01/01 50.0%
01:04:50 : 02/02 25.0%
01:04:57 : 02/03 50.0%
01:05:11 : 02/04 68.8%
01:05:20 : 02/05 81.3%
01:05:35 : 03/06 65.6%
01:05:47 : 04/07 50.0%
01:06:24 : 05/08 36.3%
01:06:50 : 06/09 25.4%
01:07:19 : 07/10 17.2%
01:07:29 : Test finished.

----------
Total: 7/10 (17.2%)


Slight hiss @ 250k and at 300k i struggled.

The bitrate for the track :

Lossywav flac ( 369 / 381)
Wv lossy (256 / 313)

[halb27]

Thank you for your tests, shadowking.

Another user - unfortunately I forgot who it was - also found that --portable wasn't transparent with a specific sample.
That's why a more conservative preset scheme was introduced, which ATM has to be explicitly invoked by the --altpreset (or -t) switch.

Do you mind trying your sample using --portable --altpreset?

Other than that I'm not quite sure about your lossyWAV test methodology. Sounds a bit as if you did 5 trials for each track and queued the results for both tracks as a 10 trials result.
I think 8 trials should be done at least for each track, and the results shouldn't be mixed up.
To cover fatigue you can of course do two 5 trial tests with any of your samples and gather the results for this sample in a 10 trial list. This procedure is a 10 trial test for the sample, with a long pause between the 5th and 6th trial.

This post has been edited by halb27: Jan 26 2010, 19:18

[shadowking]

I did a total 10 trials per track and took a little rest at 5. Is Q2.5 in v1.2 same as previous versions ? I think Q2.x yields great results anyway. Quality 3 looks interesting as bitrate is half of lossless but higher than -P. Nick said extra triggers kick in at that setting. Unless one is very paranoid about bitrate , Why not quality 3 for additional headroom ? Which one is safer - Q3 or Q2.5 -altpreset ?


To make it clearer: I realize that below Q5 there is a very small risk of a deviation in noise and it doesn't bother me. If Q2 is already good my guess is Q3 will be very close to true transparency, very hard to abx which is where i want archival level encoding at.

This post has been edited by shadowking: Jan 27 2010, 01:31

[shadowking]

Okay. I went strait to Q2.5 --altpreset, 10 trials without warming up.

foo_abx 1.3.4 report
foobar2000 v0.9.6.8
2010/01/27 21:13:47

File A: C:\windows\profiles\ng\My Documents\temp\Ivri Lider\The New People\09 - Dear Sir.flac
File B: C:\windows\profiles\ng\My Documents\temp\09 - Dear Sir.lossy.flac

21:13:47 : Test started.
21:14:29 : 01/01 50.0%
21:14:58 : 01/02 75.0%
21:15:14 : 02/03 50.0%
21:15:36 : 02/04 68.8%
21:16:00 : 03/05 50.0%
21:16:27 : 04/06 34.4%
21:16:57 : 05/07 22.7%
21:17:40 : 06/08 14.5%
21:18:24 : 07/09 9.0%
21:19:12 : 08/10 5.5%
21:19:22 : Test finished.

----------
Total: 8/10 (5.5%)

[halb27]

Thank you for your test, shadowking, great work.

If your time allows for it, it would be great if you could find the lowest -q value that makes your sample transparent, if possible both for the default and the --altpreset quality scheme.
And it would be great as well if you could upload (or give a link to) the sample.

[Steve Forte Rio]

here is example:
lossyTAK.bat

Thank you!!

[shadowking]

Thank you for your test, shadowking, great work.

If your time allows for it, it would be great if you could find the lowest -q value that makes your sample transparent, if possible both for the default and the --altpreset quality scheme.
And it would be great as well if you could upload (or give a link to) the sample.

Sample provided:

http://www.hydrogenaudio.org/forums/index....showtopic=78244

At 6 ~ 7 secs - 'rotzehhh' has a blanket hiss. You can hear it clearly at lower Q setting


Abx @ Q3 - was too hard.

foo_abx 1.3.4 report
foobar2000 v0.9.6.8
2010/01/28 00:36:26

File A: C:\windows\profiles\ng\My Documents\temp\Ivri Lider\The New People\09 - Dear Sir.flac
File B: C:\windows\profiles\ng\My Documents\temp\09 - Dear Sir.lossy.flac

00:36:26 : Test started.
00:37:05 : 01/01 50.0%
00:37:21 : 02/02 25.0%
00:38:15 : 02/03 50.0%
00:38:22 : 03/04 31.3%
00:39:25 : 03/05 50.0%
00:39:36 : 03/06 65.6%
00:40:21 : 04/07 50.0%
00:40:30 : 05/08 36.3%
00:40:40 : 06/09 25.4%
00:40:59 : 07/10 17.2%
00:41:05 : Test finished.

----------
Total: 7/10 (17.2%)

This post has been edited by shadowking: Jan 28 2010, 11:50

[Agent69]

Very interesting Shadowking. In my testing of LossyWAV at -P -t, a LossyWAV setting which I pilfered from Nick's signature, I haven't noticed my music sounding different. So either I am blessed with tin ears or I just haven't noticed it yet (maybe both).

[shadowking]

Very interesting Shadowking. In my testing of LossyWAV at -P -t, a LossyWAV setting which I pilfered from Nick's signature, I haven't noticed my music sounding different. So either I am blessed with tin ears or I just haven't noticed it yet (maybe both).

I must stress that Q2.5 -t is not easy at all to abx , describe or explain. At lower settings i was sure in moments of concentration (q2 ~ 2.5) .Q3 felt 'impossible'. Even now i find the whole thing odd, But if it was one whole test (Q2..2.5..2.5 -t) abx would = 25/30

[halb27]

Thanks, for providing your sample, shadowking, and for the hint towards the issue. I'll try if I can hear the issue as well (not today cause I'm not totally well ATM).

Thanks also for testing -Q3. From your impression, as 7/10 doesn't really say something especially when looking at the log: do you think -q 3 is transparant, or do you think it isn't totally transparent, but you weren't able to ABX it?

It sounds fine however that -P -t seems to be very close to transparency, more or less the target of the portable quality level (though not totally satisfying from what we thought the -P quality is delivering, but it's good that you proved this wrong).

BTW what do you mean by saying 'even now I find the whole thing odd'?

This post has been edited by halb27: Jan 28 2010, 19:18

[Nick.C]

Shadowking,

Thanks for your continued effort in ABXing lossyWAV output. I am pleased that --quality 3 is proving to be so difficult to ABX.... Out of interest, was the --quality 3 processing with or without --altpreset?

Nick.

[halb27]

I tried your sample, shadowking.
At -q 1 --altpreset I oould hear the issue (though I don't now if I would have noticed it if I hadn't known the spot).
At -q 2 --altpreset I still beleive I could hear it, but wasn't able to ABX it.
Well, probably my ears aren't good enough (I celebrated my 60th anniversary last year), or maybe I didn't try hard enough.
Anyway I think the issue can be classified as very subtle at -P --altpreset, but please tell us if you disagree, shadowking.

Nick, what do you think about skipping the old preset quality scheme?
With the new scheme the quality keeps closer to that of -q 5 for all the quality settings. -P takes profit of it qualitywise, and it is even more relevant for the lower quality settings the quality of which doesn't deviate too much from that of -P.
Two quality schemes aren't attractive in the long run, and IMO the new scheme is more attractive.
Maybe we should have an intermediate named quality level between --portable and --standard, corresponding to say -q 3.5 --altpreset as a proposal, now that we know -P isn't transparent. This quality level should produce transparent results with a small safety margin, so '--transparent' can be an adequate name.

This post has been edited by halb27: Jan 29 2010, 21:46

[pdq]

(I celebrated my 60th anniversary last year)

You mean birthday. 60th anniversary would make you my parents' age.

[Nick.C]

Two quality schemes aren't attractive in the long run, and IMO the new scheme is more attractive.
Maybe we should have an intermediate named quality level between --portable and --standard, corresponding to say -q 3.5 --altpreset as a proposal, now that we know -P isn't transparent. This quality level should produce transparent results with a small safety margin, so '--transparent' can be an adequate name.

I agree that two quality schemes are not desirable and would be happy to remove one of them in the next release. However, I would have to be feeling particularly brave to call a preset "--transparent".... I'll have a think and try to come up with a possible name for the parameter. .... oh - I would probably prefer -q 3.75 rather than -q 3.5 as the quality component of the preset.

[halb27]

I see, you prefer -q 3.75 because it's midway between 2.5 and 5.0. As the exact value is a bit arbitrary anyway, this is a plausible choice. And a tiny bit more on the safe side.
I agree, '--transparent' wouldn't be a correct name as we can never prove this, '--expected-to-be-transparent' would be correct, but isn't a practical name, of course.

[halb27]

(I celebrated my 60th anniversary last year)

You mean birthday. 60th anniversary would make you my parents' age.

Yes, birthday. I thought talking about anniversary is the same thing.

[shadowking]

Sorry i was tied up with a lot of work so i wasn't on the forum.

@halb27: Your right. Its subtle and i had to learn it @ lower settings just like wv lossy. I believe it exists @ Q2..2.5 and VERY hard to tell @ q 2.5 -t. I would say Q3 is fully transparent and had to force myself to complete that test.


@nick.c : I used just normal -q3