PHONETIC LABEL (APSEG) SEQUENCE SEARCH SYSTEM AND ACOUSTIC-PHONETIC VARIATION CLUSTERING BASED ON A NETWORK REPRESENTATION.

This paper presents a search system for a finely labeled speech data base, which we have been constructing. When a phoneme sequence is specified in this system, all the APSEG label sequences corresponding to the specified one can be obtained and a network consisting of the nodes of the APSEG labels from these APSEG sequences is composed. To apply this system, we have investigated the relationship between the APSEG label durations and the pronounced lengths of a word. In the latter part of the paper, we propose a graph theoretic approach to network clustering as a method for sorting the acoustic-phonetic variations of speech pronunciation.