A real‐time pitch extraction and four‐tone recognition system for mandarin speech
Kai‐Cheng Chang, Chien‐Chiao Yang · Journal of the Chinese Institute of Engineers · 1986
Mandarin is a type of tone language, i. e., the tones are part of the phonemic make‐up of words (1). Basically, there are four tones in Mandarin. On the basis of the fact that the four individual tones in Mandarin have almost fixed patterns on their pitch variations, we have designed a real‐time, speaker‐independent pitch extraction and four‐tone recognition system for isolated Mandarin speech: The pitch periods are extracted from the zero‐crossing interval sequence of speech; and the recognition algorithm is derived from the analysis of pitch contours of the four tones. The system has been implemented on a microcomputer development system MDS‐286. The average four‐tone recognition rate for the digits “0” to “9”, uttered 10 times by 5 males and 5 females, is 90.2%. The response time is about 1.64s for one digit. Both the recognition rate and the response time can be further improved.