I wonder if any of you has compared the differences when coding from, say, a 24 bit sample to some lossy audio codec compared to encoding from a 16 bit sample?

Would it for instance be possible to, when coding from a superior resolution 32 bit sample to OGG, MPC, MP3, etc..., to get a better sample than lowering the resolution from 32 bits to 16 bits (on the wave), using far less space?

If not, why!?


I'm sorry if this topic has been debated before, but if it has, could someone point me to a link?


Erlend

I may well be wrong, but I get the impression that all the lossy codecs do is convert 24/32 bit data down to 16 bit before processing them, so unless you used reasonable settings when converting 24 -> 16 bit yourself, there shouldn't be any difference?

Does anyone who'se not guessing no the answer?

Most of lossy encoders/decoders operate on 32bit float data internally (some decoders are 16bit, like FhG MP3 decoder appears to be)

It should not improve the sound, because lossy encoders discard information that are defined inaudible by the psycoacoustic model used, and information provided by more-than-16 bits sources should be defined inaudible.
I said "should", but some programmers here can tell us if a lossy encoder will encode a 1-bit-out-of-16 peak to peak signal (-96 db).

Same thing for 96 or 192 kHz sources. Any extra information provided is immediately recognized as inaudible, and discarded.

Is there any codec that adaptively chooses bit depth on a sample? I would guess that an codec would use this kind of information to get the best quality from the file, in a addition to all the othe stuff.

It was this that triggered the question.

And if it's true that the quality may not be improved this way, would it in theory be possible to construct a codec that would do such a thing?