lossyWAV 1.2.0 released, Added noise WAV bitdepth reduction method Options

[halb27]

Just an idea about how to name the intermediate quality level -q 3.75 (--altpreset): --economic (as a short term for: a quality demand very close to that of --standard, but with a more economic bitrate).

[halb27]

Average bitrate for my standard test set of various old and new pop music:

-P --altpreset: 379 kbps
-q 3.75 --altpreset: 415 kbps
-S --altpreset: 445 kbps

Looks nice to me.

[johnb]

Could you please answer this for clarification (on my side):

For a given quality level (e.g. --standard), is --altpreset aiming to be more conservative quality-wise or is it aiming at reducing the required bitrate without perceived quality loss?

As for me, I currently use --standard for archiving (and possibly for later transcoding to lossless e.g. lame -V4) and mpc -standard for my DAPs.

Thanks for your answers.

Regards
johnb

This post has been edited by johnb: Jan 31 2010, 15:50

[halb27]

There's two things coming with the --altpreset quality scheme:

a) For quality levels up to --standard --altpreset makes lossyWAV behave more conservative (for --standard it's nearly the same, difference is the more essential the lower the -q value). Above --standard lossyWAV reduces the overkill quality demand a bit.

b) The frequency limit of the noise analysis is lowered a bit compared to the old scheme (from 16 kHz to 15.2 kHz).
Noise analysis must be limited because otherwise in many situations no or low energy would be found in a tiny frequency region driving lossyWAV to keep all or nearly all of the bits for no good reason.
--altpreset has the effect here to make lossyWAV more efficient.

[gottkaiser]

Hi,

I was "playing" around with converting some tracks with lossyWAV v1.2.0 because I'd like to transcode my music library.
Two problems I noticed with the processed tracks in Winamp v5.572:

1. The basic spectrum analyzer next to the duration display is staying "flat".
2. The milkdrop visualization plug in is not usable because the graphics are not moving anymore.

They are connected and go back to the nonresponsing spectrum analyzer I guess.
I would like to use lossyWAV but I also like to use the milkdrop visualization. Any tips or thoughts about this?

[Nick.C]

Does this plug-in work with FLAC?

[gottkaiser]

Yes, it works codec independent.
I used it for years and it worked flawless. Is it possible that lossyWAV cuts certain frequencies? I'm not into that subject but I know the plugin analyses the music and generates movements depending on the music. But with the lossyWAV files it just stays "flat".

This post has been edited by gottkaiser: Feb 1 2010, 01:26

[gottkaiser]

I tried converting the lossyWAV tracks to mp3 (vbr). Then the visualization responses normal again.

[Nick.C]

How does the visualisation deal with the unprocessed WAV? Similarly, the decoded WAV from the lossyFLAC file?

[gottkaiser]

I decoded the lossyFLAC file to WAV with and without "--keep-foreign-metadata". Both files work fine with the plug-in.
So I guess the problem seems to be with Winamp.

Thanks for your help and sorry for the hustle.

This post has been edited by gottkaiser: Feb 1 2010, 17:32

[benski]

I decoded the lossyFLAC file to WAV with and without "--keep-foreign-metadata". Both files work fine with the plug-in.
So I guess the problem seems to be with Winamp.

Thanks for your help and sorry for the hustle.

This will be fixed for Winamp 5.573/5.58/5.6 (whatever the next version is It happens when FLAC frame sizes areless than 576.

[Nick.C]

Thanks very much, Benski.

[gottkaiser]

great. Thanks fro me as well!

[Northpack]

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.

How about lowering --extreme to 6.25 then and --insane to 7.5? Guess 10 is über-insane anyway... regarding that even portable is almost transparent (it certainly is for my ears). Great work btw!

[halb27]

Overkill quality demand of a minor degree can also be done by giving the -q values beyond 5 a less defensive meaning.
Nick does exactly this with the --altpreset scheme, but IMO there can be more to it.
Nick doesn't want to change the quality scale 0...10, and in fact there is no need for it as -q 10 can be configured internally to yield a significantly lower bitrate than with the old scheme.

This post has been edited by halb27: Feb 2 2010, 11:31

[shadowking]

Personally, i like the old scale as its simple and what I'm used to.

We could split into two categories for different people and their requirements :

Compact file sizes:

Q0 -- zero - lower quality , high chance of artifact
Q1 -- medium - medium quality, some chance of artifact
Q2 -- portable / compact - high quality, normally transparent (with small risk of artifact)

Larger files, suitable for archiving and transcoding:

Q3 -- standard - very high quality - transparent on most test samples.
Q5 -- extreme - Very high quality. Transparent with slight overkill
Q6 -- insane - Extreme high quality. Transparent with more overkill
Q7 -- Ultra - Extreme high quality. Transparent with lots of overkill


Now its clearer that a portable switch should belong to catergory 1 and a 'standard' or default to catergory 2.
Q1 or 2 are contenders for --portable and like wise Q3 could be the lossywav default and Q5 renamed to --extreme
However, Q5 is also okay as the default if we take a paranoid stance.


To simplify further with a few switches using my alternative scale;

--portable / compact = Q2
--normal / standard = Q3
--extreme = Q5

This post has been edited by shadowking: Feb 2 2010, 13:20

[Northpack]

Q0 -- zero - lower quality , high chance of artifact
Q1 -- medium - medium quality, some chance of artifact
Q2 -- portable / compact - high quality, normally transparent (with small risk of artifact)

I don't think that such lq presets are necessary. LossyWAV isn't designed to deliver reasonable quality at Q1 or Q0. I suggested the presets stretching from 2.5 to 7.5 because it seems to me that the existing quality scale is too wide to be covered by resonable presets - too low at the bottom, too exesssive at the top. If this is too narrow it could be like --portable 2, --economic 3.5 --standart 5, --extreme 6.5 -- insane 8. Anyway, I think Nick should know best which presets/scale-equation is most appropriate.

This post has been edited by Northpack: Feb 2 2010, 14:26

[halb27]

The 2 groups you built, shadowking, are essential IMO:
Group 1 targeting at HQ lossy encoding, group 2 targeting at an efficient alternative to lossless.
As for the 2nd group IMO it shouldn't start below -q 5, because in this area there is the definite demand for transparency, and -q 5 is what is in line with the basic lossyWAV principle.
For group 1 there is a wide variation of quality demand what people think is acceptable. At the moment there is just 1 named quality level here (--portable), but a more stronger one (call it --economic or whatever) is in discussion.
With the --altpreset scheme there is room for a named quality level below --portable, too (call it ultra-portable or whatever).
What exact -q values to correspond with the named quality levels is a matter of taste of course, but what's done so far is quite alright for me, especially if we should get a namerd quality level below --portable.

This post has been edited by halb27: Feb 2 2010, 15:55

[shadowking]

A slight revision taking into account your input;

Category 1 - smaller files HQ lossy [ quality 2 ~ 4 ]

--compact / economy -Q2
--portable -Q2.5 -t
--high -Q3 .. 3.75

or:

--portable1 - p1 = q2
--portable2 - p2 = q2.5t
--portable3 - p3 = q3.x


Category 2 - lossless alternative [ quality 5 ~ 7 ]

--standard -Q5
--extreme -Q6
--insane -Q7

This post has been edited by shadowking: Feb 2 2010, 17:01

[GeSomeone]

Thanks all involved for the listening tests.

My 2ct, for what it's worth:
I exclusively use the --altpreset scale, I never use lower than -q 3 (when the --impuls setting first was introduced I thought it was needed but I'm sorry not to back that up with ABX evidence). I use always the -q and not the names for presets.

You can come up with whatever names you like, just a suggestion: don't put to much meaning in those names, except where they are on the scale (low, standard-low, standard, standard-hi, hi) or something like that. What to use it for, is more or less a personal preference. (Even if it's transparent or not can be different from person to person)

Group 1 targeting at HQ lossy encoding, group 2 targeting at an efficient alternative to lossless.

It is the same thing really.

[halb27]

Group 1 targeting at HQ lossy encoding, group 2 targeting at an efficient alternative to lossless.

It is the same thing really.

I should have been more precise:
Group 1 targeting at HQ lossy encoding for listening purposes, group 2 targeting at an efficient alternative to lossless for archiving purposes.

This post has been edited by halb27: Feb 2 2010, 23:02

[shadowking]

I decided to try again today .

q1.5 - 7/8
2.0 - 7/8
2.0 - 8/8
3.0 - 9/10

--altpreset

q1.0 - 7/8
q2.0 - 5/10


overall --altpreset has a positive effect - but is it for all samples ?


foo_abx 1.3.4 report
foobar2000 v1.0
2010/02/14 22:36:44

File A: C:\windows\profiles\ng\My Documents\music\abx\ha\submit\09 - Dear Sir-sm.flac
File B: C:\windows\profiles\ng\My Documents\temp\q3- 09 - Dear Sir-sm.lossy.flac

22:36:44 : Test started.
22:36:56 : 01/01 50.0%
22:37:04 : 02/02 25.0%
22:37:16 : 03/03 12.5%
22:37:32 : 04/04 6.3%
22:37:58 : 04/05 18.8%
22:38:57 : 05/06 10.9%
22:39:28 : 06/07 6.3%
22:40:04 : 07/08 3.5%
22:40:16 : 08/09 2.0%
22:40:55 : 09/10 1.1%
22:45:32 : Test finished.

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


foo_abx 1.3.4 report
foobar2000 v1.0
2010/02/14 22:50:01

File A: C:\windows\profiles\ng\My Documents\music\abx\ha\submit\09 - Dear Sir-sm.flac
File B: C:\windows\profiles\ng\My Documents\temp\q2-t- 09 - Dear Sir-sm.lossy.flac

22:50:01 : Test started.
22:50:35 : 01/01 50.0%
22:50:41 : 02/02 25.0%
22:51:02 : 03/03 12.5%
22:51:11 : 03/04 31.3%
22:51:23 : 03/05 50.0%
22:51:33 : 04/06 34.4%
22:51:42 : 04/07 50.0%
22:51:49 : 04/08 63.7%
22:52:24 : 05/09 50.0%
22:52:41 : 05/10 62.3%
22:52:47 : Test finished.

----------
Total: 5/10 (62.3%)

[shadowking]

Also playing around with this interesting new --altpreset. I tried more samples and Q1 -t is hard to abx and bitrate is similar to old q2.

This post has been edited by shadowking: Feb 14 2010, 14:24

[halb27]

... overall --altpreset has a positive effect - but is it for all samples ? ...

Nobody knows for sure, but it should be like that.
Quality demand is higher below -q 5, very significantly higher at the very low quality levels so inappropriate results for lossyWAV shouldn't exist even for low -q values.
At the --portable quality level general quality demand is still higher compared to the old scheme though bitrate is more or less the same.
The bitrate saving feature comes from the restriction of the noise analysis which stops at roughly 15.1 kHz compared to 16 kHz with the old scheme.
Low energy in small frequency areas in these high frequency regions drives the lossyWAV mechanism to save only few bits if at all for no good reason.
So in a sense the --altpreset variant takes care of quality for the better reasons - at least it is expected to do so.

This post has been edited by halb27: Feb 14 2010, 15:24

[2E7AH]

If I could edit my posts, I would add this warning on posts #38 and #57:

Correction file won't be created if filename name is Unicode (or contains character not in user code page): link