treevq - build/use a binary tree codebook for vector quantisation


treevq (-i item|-a item) (-c cbitem) (-f cbfile) (-n numcode) (-u) file


treevq is a program to either build or use a binary-tree format vector quantisation codebook. In encoding mode (when -c and -f options not used) the input item(s) are divided into 2 clusters, then each cluster may then be split into two further clusters, etc, until the number of requested codes is met or exceeded. The cluster centroids used in the derivation of the final clusters, and the final clusters themselves are then saved back to the file. In decoding mode (where -c or -f is used) a built codebook is used to encode a single new input item, the result saved as an artificial coefficient item consisting of the codebook vectors, but with the 'flag' field in each coefficient record set to the mapped codebook index. For decoding, the codebook may be stored in a separate file. For encoding, a uniform (balanced binary tree) may be enforced with the -u switch.


-I Identify program name and version number.

-i item Select input coefficient item for building or encoding.

-a item Select all coefficient items in the file for building.

-c cbitem Select input codebook coefficient item. Also selects encoding operation. Default: building.

-f cbfile Select input codebook filename. Default: input file.

-n numcode Specify minimum # codes in codebook. Default: 64. For -u switch, the number of codes will always be a power of two.

-u Specify a uniform binary tree. This gives slightly worse distortion for a given codebook size, but gives the best average encoding speed.


COEFF Input vectors for building or encoding.


COEFF (Building)Codebook.

COEFF (Encoding)Encoded utterance.


input= Input items for codebook.

size= Number of codes

uniform Uniform binary tree.


1.0 Mark Huckvale


<associated programs>


Fri Jul 09 14:54:38 2004