First International Conference on Spoken Language Processing (ICSLP 90)

Kobe, Japan
November 18-22, 1990

Speech Recognition Method Based on the Dual Processing Nature of Speech Perception

Hyung Soon Kim (1), Chong Kwan Un (2)

(1) DigiCom Institute of Telematics, Seoul, Korea
(2) Communications Research Laboratory, Department of Electrical Engineering, Korea Advanced Institute of Science and Technology, Seoul, Korea

In this paper, we propose a speech recognition method based on the dual processing nature of speech perception, where vowel-like sounds and nonvowel-like sounds are processed differently. For this purpose, vowel-like portions and nonvowel-like portions of the speech signal are segmented first, and different distance measures are applied to each of them. In order to segment input utterance into vowel-like and nonvowel-like intervals, we also propose a segmentation algorithm based on a set of descriptive features derived from the auditory filter bank output. According to our recognition experiments, a reduction of over one third in recognition errors is possible with the proposed method in comparison with the conventional method. This method is particularly useful when a broad phonetic classifier is employed as a front-end stage for reducing the number of candidate words.

Full Paper

Bibliographic reference.  Kim, Hyung Soon / Un, Chong Kwan (1990): "Speech recognition method based on the dual processing nature of speech perception", In ICSLP-1990, 257-260.