I'm between lossy codecs at the moment. Was using mp3, and until I get my next DAP I can't decide between vorbis and itunes. So I've decided that until my next DAP, I'll use Wavpack lossy and transcode from that later. But I'd like to test several samples to find my target bitrate.

I've read descriptions that Wavpack coding just adds quantization noise rather than traditional psychoacoustic-related artifacts. So if I'm looking for added noise, would the easy "killer" samples be quiet, classical-type samples where added noise would be more noticeable? Or any kind of isolated instruments/sounds? Or am I barking up the wrong tree?

edit: wasn't sure what forum to put this in since it's not really a lossless question -- please move it to whereever it should go.

The few problem samples for wavpack & dualstream are more like 'anti-music' - rare isolated single sounds. It is hard to find killer samples so I wouldn't try or concentrate too much on that. I test solo instruments, flute, bagpipes, voices, violins , pulsating electronic , any stuff with 'space' - these are the more critical samples.

Pretty much anything that would be on CD's that you buy should be transparent by 350k. At 250k these tonal samples frequently contain added hiss, usually subltle and not annoying but its there in abx mode with volume pumped up. By 300~350k the hiss either 'dissolves' in the music or is so subtle that its hard to even abx.

Wavpack has noise shaping and mid-side stereo enabled by default. This can reduce hiss, but the noise shaping can degrade other samples as its not adaptive (Bryant has some auto-NS solution in a future release). Joint stereo can give slightly better quality, but it can also degrade quality samples with limited channel correlation. Using -x enables smart joint stereo. The -j switch forces a given stereo mode. -s0 can turn off noise shaping (do not mess with other values). Using -h increases quality and also encode / decode time, but I would highly recommend it and its still pretty fast.

Some examples:

-hb320x
-b320x
-j0s0320 (my custom line)

http://wavpack.com/wavpack_doc.html

The other encoder you should seriously look at is Dualstream. It is has a vbr quality mode, no joint stereo. You should be hitting transparency at slightly lower bitrates than wavpack - around quality 2. The is an auto noise shaping option with --ans that can correct problems, but in some situations it can also worsen some samples, but its still better than the current wavpack implementation. You can also use a speed hack without affecting quality at all:

--encode --mode fast --optimize none --quality x.xx

http://losslessaudio.org/DualStream.php

Thanks Shadowking. That helps a lot. Samples with space is what I meant about isolated sounds. I think I may leave joint stereo on since I think most of my stuff is probably more channel-correlated than not.

I just found your thread about noise shaping, and it's very interesting. Is 4EXP the experimental encoder that disables noise shaping by default?

In the 4exp an adaptive NS is always enabled.