Skip to main content
Topic: Wavpack 3.98 auto noise shaping (Read 3578 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Wavpack 3.98 auto noise shaping

This post has been about 4 days in the making, in that I keep starting it, and then running out of time, so here we go again...

After spending time on some other stuff, like a LITTLE listening test for Roberto, I've been checking out David's new Wavpack 3.98 with auto noise shaping, and here are my thoughts.

I took both Wavpack 3.98, plus the b - release of Optifrog Dualstream and did some listening tests with each at 275 and 320 kbits, with and without noise shaping, but my main interest is with the auto noise shaping, with Wavpack 3.98.

Firstly, Bad Man's Song intro by Tears for Fears. Sample is here.

ofs 275 --ans, focusing on the section 3.4 - 11.0 seconds. ABX, 14/15. Missed it the first test, but started picking up a slight noise jumping in and out around the hihats. In reality still sounds very good, but it is there. Makes the hihats sound slightly "dusty".

wavpack 275, no noise shaping, ABX 15/15. Quite easy. Constant, but subtle background blanket hiss. It isn't horrible, but it makes the Wavpack version very easy for me to pick. Hard to get past it to check for other artifacts.

wavpack 275 -sa. ABX 13/15. WOW, big difference with auto shaping on. Much of the background hiss disappears into the mix. Listening carefully, there is still some there, but it is not constant. Huge improvement. The main difference to ABX apart from some noise still there is a slight change in the sound of the drums. The pitch, and immediate decay is not quite the same. Difficult to describe, but I can pick it. This slight change in the nature of the drum sound is also evident without noise shaping, but I didn't get to it because I was ABXing the more obvious hiss first anyway.

ofs 320 --ans, ABX 13/15. Very difficult and had to play around a bit to find anything, but again very slight changes in noise still jumping in and out of the hihats. If listening to cold, would describe as transparent, but as an ABX test, can still hear it, particularly directly after listening to the same at 275.

wavpack 320 -sa, ABX 10/15, 14/15. Nearly missed it, but there is still a slight change in the pitch or tone of the drums. Again, would not pick it in isolation, but only against the original wav.

wavpack 320, no noise shaping. Slight blanket hiss in the background, similar to 275 except that it is quieter.

With this sample -sa has a real benefit, in that it appears to break up the background hiss that I normally pick up with Wavpack lossy 3.97. It is still there at times, but because the auto noise shaping is spreading it around, it is not so constant. Good stuff!  B)

Ring Of Fire, by Stan Ridgway, sample is here.

I again encoded this at 275 and 320 kbits, with and without auto noise shaping.

ofs 275 --ans, ABX 15/15. Slight hiss, and noise pulsing with the distorted guitar.

wavpack 275, no auto noise shaping. ABX 15/15. Hiss quite prevalent and consistent as has been reported previously.

wavpack 275 -sa, ABX 15/15. Substantial improvement, but blanket hiss is still noticeable, although less so than before, comparable with ofs with auto noise shaping.

wavpack 320, ABX 15/15. Still slight noise present, but significantly better than 275.

wavpack 320 -sa, ABX 15/15. Some hiss still noticeable, but better than without -sa, and not as an obvious, constant background hiss. Another decent improvement.

ofs 320 --ans, ABX 14/15. Very good, but if I listen carefully, I can hear some noise pulsing around the distorted guitar still.

I have also tested -sa with much of my other music, and I have not yet found a sample where it actually messes up, which makes it much more useful than the original -s1 option from Wavpack 3.97. So far it would appear safe to use -sa all the time. In some cases, it does not offer any noticeable benefit, but at least it doesn't screw the pooch either. -s1 tended to actually add noticeable noise in some cases from my previous testing, so I usually avoided it.

There are some samples where ofs --ans works better, and others where wavpack -sa is similar or sometimes better, it depends on the situation. Generally, it would appear that Dualstream is slightly more transparent (less noise) compared to Wavpack lossy for the same bitrate, but it does depend on the sample, and -sa closes the gap some.

The only thing that stops me routinely using -sa now is that it is not supported with correction files, so I can not revert to lossless when required, which I am now using more and more.

Thanks again to David and Florin for these codecs, and like others, I am looking forward to Wavpack 4.0.  B)

Wavpack 3.98 auto noise shaping

Reply #1
Quite interesting. Maybe a little taste of what we can expect of v 4. 
Did you by any chance make a transcoding test? Could the noise shaping modify its transcodability (heh, I think I just made up this word) in some way?

Wavpack 3.98 auto noise shaping

Reply #2
Quite interesting. Maybe a little taste of what we can expect of v 4. 
Did you by any chance make a transcoding test? Could the noise shaping modify its transcodability (heh, I think I just made up this word) in some way?

I didn't directly test for transcodability (nice word!) this time, but I have previously. I was involved with some of the prerelease testing of 3.98, and I checked it out then. It transcodes very well from my experience, and the noise shaping doesn't seem to affect things in this regard.

Look out for the public transcoding test being organised, where Wavpack 3.98 -sa will hopefully be one of the conditions tested.

As for Wavpack 4, I don't want to put unnecessary pressure on David, but I reckon Wavpack 4 will kick 3.98's ass... 


Wavpack 3.98 auto noise shaping

Reply #3
Where did you get Wavpack 3.98 from ?

The website only provides 3.97, which doesn't support -sa.

Wavpack 3.98 auto noise shaping

Reply #4
Where did you get Wavpack 3.98 from ?

The website only provides 3.97, which doesn't support -sa.

Here's the thread where I describe the beta release and provide a link to the file:

Thanks for the testing!

Since this option reduces the high-frequency noise added to samples that don't have a lot of high-frequencies to begin with, it could actually help with transcoding. I can't imagine how it could make things worse, but we might learn more during the transcoding test.

SimplePortal 1.0.0 RC1 © 2008-2019