WavPack 4.0 Beta Release 2, including sources Options

[bryant]

The second beta release of WavPack 4.0 includes the following:

• stable format (beta1 files okay too; beta1 decoder obsolete)
• decodes and plays legacy WavPack files (vers 1.0 to 3.97)
• sources available with new decoding library

download win32 executables plus winamp2/5 & foobar2000 plugins
download sources

This post has been edited by bryant: May 14 2004, 08:06

[den]

OK, as promised some Wavpack lossy listening test results for those who care.

Just to be sure, I grabbed the latest version linked on HA, ie beta2.zip, and got the following results.

First, I took the kraftwerk1, Waiting, queen - another one bites the dust, rushing and porcelain samples from ff123's site (thanks ff123!), decoded the flacs into wavs, and then encoded them with the beta using -hb256, -hb256x6, -b320, -hb320, -hb320x6. I didn't record times, but each encode was lightning quick except the x6 ones, which were painfully slow. (ie the non x settings took a matter of seconds for all the samples, where as each x6 encode took minutes)

I chose the above samples as I felt like a change from the samples I've used in the past with Wavpack 3.97.

kraftwerk1:
hb256 - not bad quality, but easy to ABX 10/10. Slight change detectable in the background fuzz. Also slight distortion in the three notes that come in 2.4-3.8 secs. Not really noticeable except in an ABX situation.
hb256x6 - very little difference to hb256 case. ABX 10/10
b320 - no real difference in background fuzz problem compared to hb256. Distortion on the three notes in 2.4 - 3.8 secs gone. ABX 10/10
hb320 - can not ABX consistently. Sometimes I think I can hear a hint of the fuzz, but can't consistently pick it so it is statistically transparent. ABX scores were ever better than 7/10 and were usually 5/10.

waiting:
hb256 - some slight noise around the vocals in the opening, but damn it's hard to pick every time. I can ABX it, but only consistently getting 8/10.
hb256x6 - can't ABX. Transparent. Blah!
Relatively noisy or busy samples like this one are good for Wavpack lossy to hide its sins within. It's when you have solo instruments and gaps in the music that any added noise can become more obvious.

Another one bites the dust:
hb256 - can ABX 10/10, but must concentrate, slight change in background noise in the recording, plus a hint of "dustiness" around the percussion.
hb256x6 - can still ABX, but it is much harder, and failed first attempt (6/10), but then got it the next couple trials in a row. (10/10 and 10/10)
b320 - can not ABX consistently. Sometimes I think I can hear a slight change in the b/g noise, but not consistently. Effectively transparent.
hb320 - transparent.

porcelain:
hb256 - Can ABX (10/10), slight changes in the noise/distortion that is part of the opening samples within the music.
b320 - still there, much much more subtle.
hb320 - got me, it's transparent.

rushing:
hb256 - at 4.5 secs in where some chords kick in, there is a slight but noticeable change in noise around the chords. (10/10)
hb256x6 - no change or improvement from hb256. (10/10)
b320 - bugger, it's transparent. (6/10, 6/10 and 5/10)

So in summary? hb256 is not normally going to be transparent, but it is still very good. None of these samples were disasters at 256 kbit. I wonder if Sony's new ATRAC 256kbit is this good? (I have my doubts, but I haven't tested it yet.)

hb256x6 sometimes makes some difference, but it is not consistent, and should not be counted on.

b320 is damn good and lightning quick to encode/decode. I personally would not use it as my default, as I prefer the extra security of hb320, but b320 will be transparent a lot of the time for most samples I believe.

hb320 is my choice for now. It is still quite quick, and seems to work very well at removing those little niggly changes in the noise to my ears. I can't comment on the benefits of hb320x6, as the hb320 has been transparent on the samples I have played with so far.

From the results above, you can see that I didn't get to even listen to the hb320x6 samples in this test, as transparency was reached in each case. Besides, life is too short!

I haven't extensively tested the new Wavpack 4 lossy beta for transcoding performance, but I can not think of any reason to suspect it will have changed from the previous versions. I have started transpferring some parts of my collection to Wavpack 4 beta 2, and have not stumbled across any problem transcodes to ATRAC on my minidisc yet.

Later,

Den.