Why’s FLAC ‘the’ lossless codec? Is it future-proof(able)? Comparisons, [TOS #6: was “Future proof lossless audio/codec” in General Audio] Options

[greynol]

@ Greynol: you don't seem to get the point.

If what you were trying to say was the only development on a asymmetric lossless codec with any visibility since flac was on YALAC/TAK, then yes, you're right.

[greynol]

The compression gain of TAK over the reference FLAC implementation is something like 6%, not 3%.

Link?

[Porcus]

The compression gain of TAK over the reference FLAC implementation is something like 6%, not 3%.

Link?

(Reference) FLAC doesn't fare that well with the particular test corpus at http://synthetic-soul.co.uk/comparison/jos...sion&Desc=0 , where YALAC (now TAK) yields an almost 8 percent improvement over FLAC -8. This result is at odds with the others I have seen though, but most of us probably don't have a 'representative' music collection.

This post has been edited by Porcus: Jun 24 2012, 18:05

[IgorC]

Those are old results (2006)

The last comparison (2008)
http://www.synthetic-soul.co.uk/comparison...sion&Desc=0

[greynol]

Thanks. I see 6% there, not 8% though.

Re: Old results

IIRC, TAK has continued to sacrifice compression strength for speed since 2008. Perhaps individual collections might see improvements greater than 3%, but on average I highly doubt it.

This post has been edited by greynol: Jun 24 2012, 18:32

[IgorC]

http://www.synthetic-soul.co.uk/comparison...sion&Desc=0

The highest compression:
TAK p4m - 1290.33 MiB
FLAC -8 -Ax2 - 1328.08 MiB

1328.08/1290.33 = 1.029...
aprrox. 3% of filesize reduction

This post has been edited by IgorC: Jun 24 2012, 18:34

[Porcus]

Those are old results (2006)

The last comparison (2008)
http://www.synthetic-soul.co.uk/comparison...sion&Desc=0

That's a different test corpus than the one I linked to, which is Josef Pohm's study (result hosted by Synthetic Soul).

[greynol]

That doesn't change the fact that much has changed since then.

[IgorC]

That's a different test corpus than the one I linked to, which is Josef Pohm's study (result hosted by Synthetic Soul).

It's from 2006 while there were two new realeses of FLAC later in 2006/2007 http://flac.sourceforge.net/changelog.html


This post has been edited by IgorC: Jun 24 2012, 18:58

[Destroid]

I'm late to the discussion so I am starting with earlier points in this thread.

I think FLAC is widespread because it had a head-start and probably a decent trademark name (Free Lossless...)

Future-proof is something that is never guaranteed with any software, but since most these lossless codecs are free I remind users the terms of use in the license usually says "author is not liable for any [bad things]"

It was mentioned the high number of multichannel capability is limited. I agree that quadraphonic and A-DVD are a tiny market (in some ways just the same fraction of mass-appeal as wearing 3d glasses in movies). Also, the "each instrument in the orchestra on a separate track" was something I initially thought was interesting until the real problems of software support and updating a huge monolithic file for minor changes to one track, so it seems better to stick with mono files in this regard.

One-codec-for-all is a common quest that I was better off ditching. For me, I can easily distinguish which audio files pertain to which files/projects I work with. I say: use all of them and use them for their individual merits and rule the galaxy.

[TBeck]

http://www.synthetic-soul.co.uk/comparison...sion&Desc=0

The highest compression:
TAK p4m - 1290.33 MiB
FLAC -8 -Ax2 - 1328.08 MiB

1328.08/1290.33 = 1.029...
aprrox. 3% of filesize reduction

Well, TAK doesn't fare that well with this particular test corpus...

TAK will achieve the biggest advantage if the files are highly dynamic and if it can make use of higher predictor orders of it's filters. Both doesnt't apply for the files of this comparison. While the quite old version used in this test could use up to 256 predictors, the files rarely benefited from more than 32.

This comparison rather shows the lower bound of TAK's possible advantage.

The official FLAC Comparison contains more files where TAK can use it's stronger filters:

Tak 1.0.3b (insane max): 50.60%
FLAC 1.2.1 (-8): 53.36%

This is quite close to the results of my own music collection. Which surely isn't representative.

But it possibly illustrates, that the results of Synthetic Soul's comparison hardly define the upper bound of TAK's possible advantage.

IIRC, TAK has continued to sacrifice compression strength for speed since 2008. Perhaps individual collections might see improvements greater than 3%, but on average I highly doubt it.

Well, it's a bit more difficult. Newer versions don't use more than 160 predictors (vs. 256 or more in earlier versions). This way they will loose some compression for those files, where TAK's advantage was the biggest. In exchange average compression improved for those files, where TAK's advantage was small. Overall the average compression has improved for my test sets.

Those are old results (2006)

The last comparison (2008)
http://www.synthetic-soul.co.uk/comparison...sion&Desc=0

That's a different test corpus than the one I linked to, which is Josef Pohm's study (result hosted by Synthetic Soul).

I don't think Josef Pohm's comparison should be taken into account. It contains many files from lossy sources what leads to some very specific results.

This post has been edited by TBeck: Jun 24 2012, 19:20

[greynol]

Personally I only care how comparisons fare against each other on my collection or a collection like it. Short of this, I'm only interested in average results instead of cherry-picking those that make one codec look much better or worse than another.

2-3%, not 6-8%.

[IgorC]

But it possibly illustrates, that the results of Synthetic Soul's comparison hardly define the upper bound of TAK's possible advantage.

My results are in line with Synthetic Soul's comparison.


Max. compression:
FLAC 1.2.1 -8
TAK 2.2.0 p4m

The results for 16 albums:
14 albums from here http://www.hydrogenaudio.org/forums/index....st&p=761670
15. Pink Floyd - Division Bell (rock)
16. Lady GaGa -The Fame Monster Uncensored (pop)

Bitrate:
FLAC - 872 kbps (5,798,731,761 bytes)
TAK - 843 kbps (5,605,149,806 bytes)


FLAC/TAK = (5,798,731,761 bytes)/(5,605,149,806 bytes) = 103.45% of filesize.

[Porcus]

many files from lossy sources

Now that's a good argument.

[greynol]

FLAC/TAK = (5,798,731,761 bytes)/(5,605,149,806 bytes) = 103.45% of filesize.

I was going by improvements in compression against the original instead of against each other. With this in mind I can see how a 6% improvement can grow to 8% and that the higher number is more meaningful.

Not feeling terribly sharp today.

[Porcus]

@ Destroid: a few good points, although I don't think the headstart can explain why FLAC got the upper hand on the .ape and the .wv, which were earlier.



Apart from that, at the risk of running off-topic:

Future-proof is something that is never guaranteed with any software, but since most these lossless codecs are free I remind users the terms of use in the license usually says "author is not liable for any [bad things]"

That is common regardless of cost, regardless of open/closed source, and regardless of libre/non-libre license.

One-codec-for-all is a common quest that I was better off ditching. For me, I can easily distinguish which audio files pertain to which files/projects I work with. I say: use all of them and use them for their individual merits and rule the galaxy.

Myself, I use WavPack for files ripped from CDs with pre-emphasis – simply because it is different. Yes they are tagged, they have a unique glyph at the end of the file name, but they also have a different file format. Because I want them to be easily distinguishable even if I should accidentally hit some wrong buttons in MP3Tag.

[Dario]

Well, I don’t know what kind of albums you guys use to compare those compression ratios.

I listen mostly to neofolk, and the results I get are quite different than the ones you do. Here’s the compression for the album “Forlatt” by Vàli (if you’re into good acoustic music, you may want to check it out yourself):

reference libFLAC 1.2.1 20070917 with -8: 157.129.638 bytes
TAK 2.2.0 with -p4m: 128.642.073 bytes

do you know how much difference that is (tip: it’s over 20%)? If you don't believe me, I can upload the album so you can check it out yourself. I hope you wouldn’t mind.

This post has been edited by Dario: Jun 24 2012, 20:22

[IgorC]

Well, I don’t know what kind of albums you guys use to compare those compression ratios.

I listen mostly to neofolk...

[Destroid]

@Porcus Ah, is true. The average user sometimes is suspicious of the connotation "free" but as you said this is another topic entirely.

In regards to test corpus, it may be of interest to note that almost all lossless codecs do extremely well with non-altered recordings of single instruments. But many mainstream CD releases since the late 90's have almost no semblance to natural acoustic/dynamic audio. I mention this purely in regards to the seemingly marginal gains using TAK over FLAC and also higher compression modes. I would have to do another single-instrument track corpus test to see if the "only 3% gain" argument withstands, and either way it doesn't resemble the majority of usage by the majority of users (which I imagine is CD's).

This post has been edited by Destroid: Jun 24 2012, 20:40

[Dario]

Well, I don’t know what kind of albums you guys use to compare those compression ratios.

I listen mostly to neofolk...

Quite funny, indeed. Allow me to say that the aforementioned album is probably the thing that fares best with TAK’s compression (by far!). On other albums I get a 5–7% difference in compression.

[Porcus]

Well, I don’t know what kind of albums you guys use to compare those compression ratios.

The tests linked to, disclose the test corpus. As they should.

Those figures of yours (wow ...) don't state uncompressed size. If you ripped from CD, then bitrate (per second) is probably a better figure to state, as the uncompressed is 1440. (If you didn't ... say, bought a 48 kHz file off Bandcamp, or something like that, then state uncompressed.)


For my own CD rips:
616 kb/s for those sorted by composer (classical/contemporary), and
925 kb/s for those sorted by performing artist (jazz/rock/metal ... mainly the latter) – that is 50% more music per time unit!



Records, average over a physical CD (only counting CD rips):

252 for an Edith Piaf compilation which was auto-tagged as Voice of the Sparrow, disc 2 (those things show up with various titles, this was purchased as a single CD, I don't have disc 1 nor 3). Luckily it didn't show up any CD with a long silence in there.

1345 for Merzbow: Venereology. Listen to it and understand why. WavPack x6 improves it five percent to 1275 (just to compare, it compresses my only DTS CD to 1251).

This post has been edited by Porcus: Jun 24 2012, 22:01

[IgorC]

I have encoded two sets of samples from two last public tests.
Those can be considered as independent ones.

And, yes, TAK has an advantage of 6% comparing to FLAC for both sets.

[IgorC]

I listen mostly to neofolk, and the results I get are quite different than the ones you do. Here’s the compression for the album “Forlatt” by Vàli (if you’re into good acoustic music, you may want to check it out yourself):

reference libFLAC 1.2.1 20070917 with -8: 157.129.638 bytes
TAK 2.2.0 with -p4m: 128.642.073 bytes

do you know how much difference that is (tip: it’s over 20%)? If you don't believe me, I can upload the album so you can check it out yourself. I hope you wouldn’t mind.

Yes, it will be interesting to see this kind of music which produces such a big variation of compression efficiency per codec.
30 seconds will be enough.

Thank you.

[bryant]

I have to say though, that I don't really understand why WavPack didn't catch on, being so early available. Though I don't know what features it had back in last century.

Maybe this post will at least partially answer your question.

Myself, I use WavPack for files ripped from CDs with pre-emphasis – simply because it is different.

This made me smile...it's got to be the most unique reason for using WavPack!

[TBeck]

Personally I only care how comparisons fare against each other on my collection or a collection like it. Short of this, I'm only interested in average results instead of cherry-picking those that make one codec look much better or worse than another.

If this is a response to my post: I am not cherry-picking. I didn't say, one comparison was more representative or meaningful than the other, furthemore i tried to name some audio properties which favour TAK, what implies that this will only sometimes help.

But to rely only on less TAK-friendly file sets, would be cherry-picking too and don't provide the whole picture.

Other than that i totally agree. Given the immense difficulty to create a representative comparison, the best advice for everyone probably is to try it himself on his collection.

I have encoded two sets of samples from two last public tests.
Those can be considered as independent ones.

And, yes, TAK has an advantage of 6% comparing to FLAC for both sets.

Nice, but: The sets may be independend, but somehow biased. I found, that problem samples often have some of those typical properties which favour TAK. For instance TAK can cope very well with fast transients.

Therefore those samples too aren't representative. I had to learn this myself, when i arrived at hydrogen. My own music collection and my primary test set, which consists of codec test samples from rareware.org, favoured TAK and i was a bit disappointed, when i saw some of the first test results of the users.

What i like about this thread is that it illustrates how difficult it is to create a representative comparison.

I listen mostly to neofolk, and the results I get are quite different than the ones you do. Here’s the compression for the album “Forlatt” by Vàli (if you’re into good acoustic music, you may want to check it out yourself):

Yes, it will be interesting to see this kind of music which produces such a big variation of compression efficiency per codec.
30 seconds will be enough.

I second that...