Hi all,
I had an idea to use SPEEX to reconstruct the wavform to find the residual signal... for our Lossless Audio Compression. In this case I can use all the benefits of VQ and the codebook training if required for audio, LPC parameter Vector Quantization and other stuffs....
Can some one tell me about the feasibility of this experiment(s)....
I hope better(or similar) compression and more flexibility to experiment with the parameters of lossless coding...
I can start with speech to encode and decode losslessly and then vary the parameters of speex like bitrate, prediction order and so to make it work for Audio.
Please tell me if its a positive step...or if its worth trying for...?

I also mean from the above questions that does FLAC uses CELP based algorithm ??
Can someone tells me what are the parameters which the Linear Prediction part of FLAC transmits ??
I would like it to be compared to speech coding where we send pitch, gain, LPC parameters, etc.
I also want to know about the excitation signal used in FLAC while reconstructing(in prediction back to find residual) the wavform.

I can't answer your questions directly, instead I want to remind you that FLAC and Speex are Open Source / Free Software projects. That means that you can download all the source code and documentation and start hacking.

FLAC Homepage
(find detailed infos on the documentation pages)

Speex Homepage
(find detailed infos on the documentation pages)

I doubt very much this would be any good. In the lossless case, VQ is completely useless (it's good when you want to *minimize* noise, not eliminate it).

In Lossless we try to minimize the bitrate, which is mainly from the residual side. So doesnt it mean that if we try to make the residual small enough will make our lossless coding more effcient..??
Then, I want to ask whether I can use CELP to minimize the residual signal and hence get a lossless coder.... please give suggestions on this. To me evrything is foggy till now...

CELP is a model-based technique used for low-bitrate coding of speech signals. Its performance for music is quite poor in general. Do you want to create a lossless codec optimized for speech?

Current lossless audio codecs already use some prediction techniques, as described here for instance:
FLAC (format description)
LA (theory)
Monkeys Audio (theory)
TTA (theory)
All these techniques have one thing in common: they don't use a model for the signal source. The reason for this choice is that these are general purpose audio codecs, commonly used for coding music signals.

Any new ideas in this area are welcome, so just start experimenting and create a better codec!