Recognition of Syllable-Contracted Words in Spontaneous Speech Using Word Expansion and Duration Information

Abstract

This paper presents a graphical model-based approach to syllable-contracted (SC) word recognition of spontaneous Mandarin speech. Phone deletion and pronunciation reduction are two major effects for the syllable-contracted words in spontaneous speech. In this study, the syllable- contracted (SC) words selected from a collected corpus are used for pronunciation lexicon expansion to deal with the phone deletion problem. The duration information of SC words is then incorporated to cover the effect of pronunciation reduction. The graphical model is employed to rescore all possible word sequences, including the expanded SC words, to obtain the final word sequence. In the experimental results, the Mandarin Conversional Dialogue Corpus (MCDC) was used to evaluate the proposed method. Compared with the previous work, a satisfactory improvement on the performance of the proposed approach can be achieved.