Hi,

There is an discussion at http://www.minhembio.com/forum/index.php?s...view=getnewpost which i hope Josh or someone else knowledgeable could help us to resolve.

The question is basically if you can correctly describe flac, or any part of what flac does as 'lossy', 'lossy plus corrections' or not at all 'lossy'.

The argument for describing what flac does as 'lossy' or 'lossy plus corrections' is that the linear prediction used can not fully describe the signal, thus it would be 'lossy', the rest of what flac does would then be 'corrections'.

The argument for describing what flac does as not at all lossy is that only the parts of the signal that can be compressed using linear prediction is done so, the rest of the signal (or blocks) is compressed using other methods or (horrible thought) not at all. thus nothing ever gets 'lost' in that process.

Thanks in advance for any help resolving this discussion.

The more I read the more I cannot see whether those two statements can be juxtaposed.

I don't think that I can disagree with either of those statements, except:

• There appear to be three stages, not two.
• The residual signal will certainly never be lost.

Edit: In an attempt to clarify:

What is the difference between "lossy+correction" and "lossless"?
:-)

I wouldn't go that far and divide up the prediction+correction part since it's so tightly interleaved. Together they make for a reversible filter operation which is parameterized by the prediction coefficients. Since these coefficients are also transmitted you don't lose any information --> lossless.

Cheers!
SG