Just curious if there has ever been anyone here able to ABX either DualStream or WavPack at bitrates around 400kbps and up.

I use to use Monkey's Audio for all my music, but recently converted everything to FLAC because of the two Rio Karma's I have coming.

After extensive testing over the weekend, I decided Vorbis at -q6.25 (transparent to me after much ABX testing) would probably suit my needs better as far as portability is concerned (though I must add, MPC would have certainly been preferable).

Now I'm debating what to do about my FLAC files since I probably won't end up using them on the Rio. I like the format, but compared all the other lossless choices, FLAC files suffer a bit from bloat even with using -8 for compression. This adds up to gigabytes of wasted space very quickly. I suppose I could go back to Monkey's Audio, which was why I was using it before (good compression at high setting).

Now I'm playing around with DualStream and WavPack lossy. This was something I was struggling with a few months ago actually. Supposedly, because they don't use psycho-acoustic modeling, make the best choice should you ever need to transcode to another lossy format such as MP3, Vorbis, MPC, etc. And because high bitrate ofs and wv files are pretty much un-ABXable (as far as I know), are for all intents and purposes audibly lossless. So I'm trying to convince myself that using DualStream at -q6 would be just fine and that I'd be happy, and that what I get out of it is files that take up approximately half the space that my FLAC files currently do, which is a lot at the moment.

Trouble is I'm having difficulty convincing myself. Lossy is still lossy after all, even if you can't hear what's been thrown out. Speaking of which, what does WavPack lossy and DualStream throw out anyways? I can't find any info on this. Anyways, at the moment I am considering transcoding everything and having correction files made too. Then I could store those correction files elsewhere. However this is kind of a pain to maintain when you think about it. Plus it technically means I'm still using up the space in one way or another.

Well, I've babbled enough. You can see my conundrum. I know there are a couple folks here who have gone this route already, and I'm sure had to struggle with these questions too. Really hope to hear from you guys!

Depends what you are looking for. If one must have bit-perfect audio then obviously lossless is the only way. If you need a transparent lossy that also transcodes well then DS and wavpack can fit the bill.

Note that DS Q3 should be normaly transparent and that Q6 is a big headroom to make sure than nothing will ever go wrong (like a killer sample). in case that you don't want to use correction files. Noise drops with increased bitrate so the hybrids will deliver no matter what past a certain point - maybe > 450k . Sometimes in rare cases the noise can be too high and samples might be abxable at 400k . The new wavpack and DS are often transparent at 320k

Try encoding a few albums and see

A good read on this subject is the wavpack and dualstream websites.

I have managed to ABX the odd sample at 400kbit and above in the past, but haven't tried too hard in recent times. Since the introduction of Wavpack 4, I haven't come across any examples that are obvious at 320 kbits or above with everyday listening, whereas I used to stumble across the odd example with previous versions of Wavpack.

If you were to aim for 400 kbits with either of these, I'd be surprised if you could pick it from lossless.

I won't attempt to explain what they "throw out", but in essence, the most obvious problem with these encoders is that as you lower the bitrate, there is an increase in background noise, usually present as a hiss. If you wind the bitrate right down, the music almost becomes "dusty" in that you will hear noise around the notes, but this is only obvious at 256 kbit and below.

Best to check it out yourself, but I think it won't take much to put your mind at ease. I used to keep all my correction files, but I don't usually bother now, as at 320/400 kbit, I don't miss them.

Hope this helps.

Den.

Thanks for the comments guys, I really appreciate them. I plan to sit down and do some serious ABX work today on them. Hopefully I've recovered enough from my last session over the weekend, lol.

Regarding the information these codecs "throw out" (that you can never get back without correction files), I only ask because I couldn't find an explanation anywhere. In the case of lossy codecs like MP3, Vorbis, MusePack, and so on it's quite obvious. They try to emulate human hearing and toss what one shouldn't be able to hear idealy. For WavPack, the only claim I've heard is that it is like zip technology only for music, which I believe is what I read on the official site.

It's definitely going to be hard to accept. I'm the kind of guy who shoots in RAW format with my camera for anything serious (though JPEG is just fine for snapshots and the like). The logical side of my brain keeps nagging me that once the information is lost, you can never get it back unless you re-rip. Need a stick to poke in my ear, and tell it to just shut up already.

Here's what I can tell you about what I've read (board and source) about WavPack.

The information it's discarding is purely amplitude error. No transformation is done, so all processing still occurs on the time/amplitude domains. It uses predictors to predict the path of the waveform, and stores the residual to reconstruct the real waveform on decoding. So when it functions in lossy mode, it's just less precise in storing the residual, so the output sounds a bit less like the original, but in the manner of random error (i.e. noise). WavPack does have noise-shaping algorithms to distribute that, though, in attempts to make it as unnoticeable as possible.

Someone please correct me if I'm wrong. =)

Actually this is not a terribly accurate description of the process as compared to what I have read before on these forums. Apparently codecs like Mp3 and Vorbis don't throw out sound, they actually add noise, which wipes out details of the music, thereby compressing the information needed to reproduce the signal. A psychoacoustic format like this uses its model to 'hide' the noise its adding in the music where, if it was perfect, the human ear would not be able to detect it. A nonpsychoacoustic format like wavpack basically just adds white noise to the background, thereby eliminating very quiet details in the music by raising the noisefloor. I hope this helps.

edit: I think Ariakis' description was better than mine. sorry, I should have finished reading the thread before posting.