I think this whole idea of having a correction file for the lossy encodes is excellent. I could really use this feature.
Generally speaking, which "hybrid" codec performs better encoding lossy files? And how is the lossy output comparable against, let's say, mp3? Which codec is more likely to have hardware support?
If there was any hardware support for one of these hybrid codecs, my objective would be to encode lossy files at low bitrates for portable devices, and keeping the correction files in my computer for lossless playback on winamp, or something.
From my search results, i got the impression wavpack really does a good job, and hardware support for it might be close.

Firstly, I don't think any hybrid codec has hardware support, at least not yet. Secondly, for reasons I won't get into, Hybrid codecs need more bitrate in lossy mode then pure lossy codecs. Most need 300-400kbps to remain at good quality, so low bitrates are out of the question. Personally I'm quite satisfied storing a lossless backup of all my music and then using that to encode files for the iPod.

As a different encoding method is used, I would imagine different types of audio samples cause problems for hybrid codecs compared to other lossy codecs. Is this true? If so, are there any known problems samples for hybrid codecs?

If this is the case then it probably would be hard to find a hybrid codec quality level which reaches the same level of transparency as --alt-preset standard (ie. fails on a similar proportion of samples), as it would be entirely dependent on the set of samples used for testing.

I have been using OptimFROG DualStream --quality 3 which is described by Ghido at http://losslessaudiocompression.com/ as being "normally undistinguishable from the original in ABX tests", whereas --quality 5 is described as being "transparent, suitable for archiving and transcoding". My hunch is that --quality 3 is transparent for me on my equipment as I haven't heard anything unexpected. But having said that, most of the DualStream tracks I have were ripped and encoded as soon as I got the CD, and I have never listened to the original, only the DualStream encode.

If I knew of some problem samples I could test the difference between 3, 5 and the original for myself.

One off-topic question. When I play DualStream files on foobar2000 and the .ofc correction file is also available, does it use the correction file to decode to the original, or does it just use the .ofs file?

I gave wavpack a try, and i'm listening to a lossy .wv right now. It sounds pretty decent, although i don't have ABX results yet. The average bitrate is something like 265 kbps, very constant. It is obvious, however, that these lossy encodings aren't as optimized as a "true" lossy codec is.
Since FLAC is now part of the Xiph.org foundation, wouldn't it be awesome if flac evolve to a hybrid codec too, using vorbis for the lossy part? Vorbis is way more tuned than any lossy part of the current hybrid codecs, and it has hardware support.
That'd save a lot of time for people like me and kl33per, who transcode their flacs to a lossy format for portable playback. Vorbis is also natively supported by a lot of players, including winamp.

jcoalson, or any other developer, any chance of seeing this implemented in flac? (or vorbis?)

Eh, just use MPC + APE correction file...I tried it once and it works decently well :B

can u be more precise about this?
never hear about ape correction for mpc encodes...

cu.
pipo.

Basically, just decode the MPC and subtract that from the original WAV. Store the difference as an APE. Then you can recover the original by adding MPC + APE together. However, unless this were automated in some way, it's too much trouble to be practical.

A quick note about Wavpack Hybrid versus Optimfrog Dualstream:

The lossy encoding part of Optimfrog supposedly has higher quality than Wavpack, but at a performance cost. I.e. it reaches transparency at a lower bitrate, but is slower to encode.

IIRC, this was reported and verified by ghido, bryant and den.

About using correction files with MPC: tigre wrote a post about that a while ago:

http://www.hydrogenaudio.org/forums/index....ndpost&p=194036

hum... sure.
instead of going this this way i would go lossless.
thanks.

cu.

I would just add that a new version of WavPack is expected to be released soon, you cna find in the news that it is being beta tested at the moment. Although I haven't tested this personally - the lossy compression is has been greatly improved and should definately be re-evaluated against Optimfrog.

Interesting concept, but in fact that's not a very practical solution :\
An hybrid version of flac (with ogg support for the lossy part) would be perfect in every aspect, imo.

The Foobar plugin only decodes the ofs file.

Only OptimFrog-DS itself has the capability to decode and combine ofs + ofc.

But as ofs + ofc = ofr. Why not use OptimFrog Lossless ? The ofr files would be slightly smaller than ofs + ofc.

Edit : Can't spell.

I have a DAT drive in my PC and spare tapes. When I run out of disk space I can easily move the ofc files onto tape. Then if I ever want lossless (for example if I decide to change to another format) then I can easily get the ofc files without having to rerip the CDs.

At the moment I still have spare disk space, so the ofc files are still with the ofs files.


Anybody know of problem samples specific to hybrid lossy files? Or are they the same as for any other lossy encoder?

I haven't come across any problem samples specific to the hybrid codecs, most of the testing I've done has been between OptimFrog-DS, Wavpack Hybrid & lame 3.90.3.

Florin Ghido tested OFDS on about 37 Gb of audio data, so I guess he'd know better than anyone else.

I don't believe there would be many "problem samples" in the traditional sense of the word, because hybrid codecs don't work at all like transform codecs. Since the error introduced is only quantization noise, it should only be really noticable on quiet passages, or as a subtle change in the color of background noise. At least, if I remember correctly from Den's extensive listening tests.

One really bad sample I've heard, though, is keys_1644ds, at least with Wavpack Hybrid... I haven't tested it yet with the newest beta, but Wavpack also had problems with Bryant's furious sample.