Fourth European Conference on Speech Communication and Technology

Madrid, Spain
September 18-21, 1995

Automatic Recognition of Cantonese Lexical Tones in Connected Speech by Multi-Layer Perceptron

Ying Pang Ng (1), P. C. Ching (2), L. W. Chan (1)

(1) Department of Computer Science, The Chinese University of Hong Kong, Shatin, New Territories, Hong Kong
(2) Department of Electronic Engineering, The Chinese University of Hong Kong, Shatin, New Territories, Hong Kong

In automatic tone language recognition, lexical tone identification forms an integral part of the recognition process. In this paper, automatic recognition of Cantonese lexical tone in connected speech is tackled. Cantonese is a commonly used Chinese dialect and is very rich in tones. Data for this study are from a male native speaker of Cantonese. The study proceeds in 3 stages. Stage 1 studies the feasibility of lexical tone recognition using manually segmented F0 contours. Stage 2 is concerned with the design of a novel approach in automatic segmentation of F0 contours. Stage 3 develops a lexical tone recognition algorithm using automatically extracted F0 contours. Preliminary results show that for a fully automatic system, a tone recognition rate of 72.1 % (N=725) is achieved for connected speech.

Full Paper

Bibliographic reference.  Ng, Ying Pang / Ching, P. C. / Chan, L. W. (1995): "Automatic recognition of Cantonese lexical tones in connected speech by multi-layer perceptron", In EUROSPEECH-1995, 2205-2208.