Has anyone ever considered using a hybrid approach to audio compression, where multiple techniques are used to achieve the best quality at a given file size?

Along these lines, I was thinking it might be interesting to intersperse some losslessly compressed frames along with mp3 frames within a single VBR file (I realize it would break the format and would no longer technically be an MP3 file)... but theoretically, is this something that's ever been considered?

Without knowing a whole lot about how an encoder like LAME figures out when to use higher bit rates in VBR encoding, I'll assume that it has a reasonably good (and constantly improving?) algorithm for figuring out when a passage is complex and is at risk for non-transparency. Well, what if instead of only being able to go to 256 or 320, it could go all the way to a losslessly compressed frame? It seems that if the algorithm to detect complex passages is really good, then with a small proportion of lossless frames, one could maintain transparency that even the pickiest of audiophiles can't discern from the original, with filesizes that aren't much larger than say an MP3 encoded at 320 or VBR 256.

The 320kbps limit was solved in all codecs that are more modern than MP3.
There is no need to go 100% lossless for transparent coding, especially since coding efficiency would drop to an extent that the file would be much larger.

perhaps what you are looking for is a true VBR format - something like musepack or ogg vorbis. these formats are able to scale VERY high (especially musepack) and do not have the bitrate constraints of mp3.


later

It could be done, the big question is: Why?

Since:

- even 320kbps is (apart from the very rare problem-sample) more than sufficient and easily transparent to all but perhaps five people in the world
- all modern codecs are capable of even higher bitrates
- lossless compression is IMHO used as a 'just to be sure' option and not needed for quality purposes

there is no real reason for it.