First International Conference on Spoken Language Processing (ICSLP 90)

Kobe, Japan
November 18-22, 1990

The Optimal Gain Sequence for Fastest Learning in Connectionist Vector Quantiser Design

Lizhong Wu, Frank Fallside

Cambridge University Engineering Department, Cambridge, UK

Kohonen's self-organising algorithm has been widely used for the design of connectionist vector quantisers (CVQ). One of its features is that the weight update gain sequence ?/m) is a decreasing function of the number of iterations, and if incorrectly chosen can lead to very long training times. Here we derive the time-optimal gain sequence and demonstrate its efficacy for a number of cases. It is demonstrated that the new method is time optimal and that its performance tends to that of VQ with the LBG algorithm. Finally the CVQ for linear predictive data with respect to the Itakura distance measure is applied to a multipulse linear predictive speech coder using data from the TIMIT database. Comparisons are made of waveforms and rate distortion functions.

Full Paper

Bibliographic reference.  Wu, Lizhong / Fallside, Frank (1990): "The optimal gain sequence for fastest learning in connectionist vector quantiser design", In ICSLP-1990, 1029-1032.