7th International Conference on Spoken Language Processing

September 16-20, 2002
Denver, Colorado, USA

Benefit and Cost Analysis of Using the Improved Vector Quantizer Design Algorithm for Glottal Source Waveform Compression

Peter Veprek (1), Alan B. Bradley (2)

(1) Panasonic Speech Technology Laboratory, USA; (2) RMIT University, Australia

Vector quantizer design is an essential step required in the development of many quantization (compression) and clustering (classification) tasks. The iterative codevector readjustment and reassignment algorithm (ICRRA) is shown to outperform the traditional generalized Lloyd algorithm for the task of glottal source waveform compression. Analysis of the results reveals that the ICRRA offers improved performance at a very high confidence level. The improvement is achieved at the expense of increased computation. Storage requirements of the algorithm, however, remain virtually unchanged for typical conditions. Overall, the ICRRA is particularly suitable for off-line processing where longer execution time is well justified by the benefits of reduced reconstruction distortion.

Full Paper

Bibliographic reference.  Veprek, Peter / Bradley, Alan B. (2002): "Benefit and cost analysis of using the improved vector quantizer design algorithm for glottal source waveform compression", In ICSLP-2002, 1841-1844.