Synthesis‐based recognition of continuous speech

Abstract

An acoustic phonemic recognition system for continuous speech is presented. The system utilizes both context‐dependent and context‐independent characteristics of the speech signal to achieve recognition. The interphonemic contextual effects contained in formant transitions are incorporated into the system by using the synthesis‐based recognition approach of Thosar and Rao [IEEE Trans. Acoust. Speech Signal Proc. ASSP‐24, 194–196 (1976)]. It is shown that the information contained in transition segments of the speech signal improves the performance of the system considerably. Recognition of continuous speech is accomplished here in three stages: segmentation, steady‐state recognition and synthesis‐based recognition. The system has been tried out on 40 test utterances, each 3–4 s in duration, spoken by a single male speaker and the following results are obtained: 5.4% missed segment error, 8.3% extra segment error, 52.3% correct recognition using only steady‐state segments and 62.0% correct recognition using both steady‐state and transition segments.