[vorbis-dev] VQ codebook sanity check (more musings)

[Monty]

More musings and questions for those in the know...

The fixed length codeword (FLC) and variable length codeword (VLC) codebooks
appear to usually end up with approximately the same average error.  Am I
imagining this or is it roughly correct?

If so, then my previous assumption is correct; the FLC books have better 'high
probability' precision, and worse precision in the fringes given a specific
codebook size. VLC is the opposite.

However, it's also implied that the VLC books end up using less bits!  Because
the codeword length is varied by probability of a hit (like huffman coding), we
get some savings.  So, the VLC codebook would appear to have an edge over the
FLC in terms of encoding efficiency.  Perhaps the savings in bits approximately
makes up for the loss of precision in commonly used entries over FLC? If so,
this would make VLC indesputably superior. 

Something tells me it's not that simple :-)

Monty

--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/