A VQ-based preprocessor using cepstral dynamic features for speaker-independent large vocabulary word recognition

A VQ (vector quantization)-based preprocessor is proposed which reduces the amount of computation in speaker-independent large-vocabulary isolated-word recognition. The features introduced here are the use of a universal codebook in the VQ-based preprocessor and the use of multiple feature sets including cepstral dynamic features. Word-specific codebooks are used for front-end preprocessing to eliminate word candidates whose distance scores are large. A dynamic time-warping (DTW) processor based on a word dictionary, in which each word is represented as a time sequence of the universal codebook elements (SPLIT method), then resolves the choice among the remaining word candidates. Recognition experiments using a database consisting of words from a vocabulary of 100 Japanese city names uttered by 20 male speakers confirmed the effectiveness of this method. The total amount of calculation necessary in this condition is almost 1/10 of that without preprocessing