Abstract

A phoneme-based Gaussian mixture VQ codebook can improve the conventional DHMM system performance significantly. In this paper, an optimization method for the phoneme-based VQ codebook is proposed. The experimental results shown that the optimized phoneme-based VQ codebook leads to both the improvement of system performance and the reduction of system complexity. INTRODUCTION In a phoneme-based Gaussian mixture codebook [1], each phoneme of each sex of speakers in the training sequence is represented by a codeword. This means that we treat the phonemes as data classes in the clustering by VQ. The number of codewords in a codebook, codebook size, is normally larger than the number of phonemes in the training sequence. The reason for this is that the phoneme space is only a subspace of the speech space. In other words, the speech signal space includes not only the phonemes, but also other sounds between the phonemes and silence or background noise. We generate a Gaussian mixture codeboo...

Citations