I have a rather large collection of MiniDisc recordings captured as 44.1 kHz/24-bit/Stereo. Most of them originated years ago from 16-bit sources, some are recorded directly from magnetic tape. Most tracks contain some noise from a mixing console.

These files are currently taking up way more space than they are worth: around 1400 kBit/s FLAC, except those manually converted to Mono. Which lossy codec would be the best choice for them?

High bitrate WavPack lossy might be the ideal solution for this. It will maintain the dynamic range of the 24-bit source and might avoid transcoding artifacts, while not using the absurd bitrates of lossless. I have used this for archiving DCC recordings (which are mpeg 2) and I know someone is using this for archiving old TAC encodings without resorting to lossless.

If you want to try, I would recommend using the new alpha version that has the dynamic noise shaping and using a command like -hb384x --dns. That should be 3-4 times smaller than lossless.

Great, I can now encode at least the non-important recordings.

Is WavPack/dns fully compatible with older decoders, in particular Foobar 0.8.3 / decoder 2.3?

Do you recommend exactly 384 kBit/s?

How well WavPack lossy would perform if I decide to transcode to the same format in the future (for example, after cut-n-paste editing)?

No matter what you've fed to your MD-recorder years ago, music on MD is stored as 44.1KHz 16bit, under every and any circumstance. Capturing the music from the MD as 24bit therefore seems a waste of bits to me.

Surely, as a transform / hybrid codec, ATRAC doesn't have any native bitdepth? The decoder can invert the transforms to any accuracy. Certainly some MD players claim to have at least 20-bit decoders and D/As. IIRC some MD recorders had 20- or 24-bit A/Ds at the input too.

It's just like mp3, which can be decoded to any bitdepth.

However, if the input was via a "noisy mixing console" then 24-bits is overkill.


Back on topic: If lossyFLAC was finished, I'd suggest that.

If you have transcoding concerns, you could use wavpack lossy in hybrid mode and dump the correction files to DVD-R and forget about them, until/unless you decided you needed them.

Cheers,
David.

The only switch that causes any incompatibility issue is --optimize-mono; if you don't use that then you're fine back to the 4.0 release. Also -hh is not recommended for hardware playback (although Rockbox iPod is the only case where it's ever been a problem). At this point, --dns will not work with correction files (i.e. it's lossy only).

I mentioned 384 because it one of the "standard" bitrates. I like to use them (256, 320, 384, 448, 512, ....) but it's not required (and WavPack lossy doesn't hold them that close anyway). I also picked that rate as a decent compromise for what you said, but it is all based on how important you feel these are. If really want to make sure there's no audible degradation and think you may re-encode them a few times, then I would go for 512 kbps. For the noisy "non-important" recordings, 320 kbps would probably be fine (assuming they're not recordings of weird electronic noises).

WavPack lossy holds up well to multiple encodings, with very little additional noise being added after a few generations. I would venture that if you used one of the bitrates I mentioned above, you could re-encode it dozens of times and it would still be better than the quality of the next lower bitrate. Caveat: I have not tried this with the new --dns option, but see no obvious reason it wouldn't be the same.

It is a click of the mouse to throw the excessive bits away later in Foobar, like I'm about to.

Thank you for your thorough explanation, Bryant. Wavpack it is then.