摘要
本文介绍了基于连续分布型HMM的汉语连续语音声调识别方法，提出了一个适合于汉语连续语音声调识别的特征参数提取和识别方案。通过对汉语连续语音声调特点的分析，选择了8个音节单位的连续分布型HMM作为声调识别用基元模型进行识别试验，识别结果表明，10名话者1070个句子的连续语音声调识别的平均识别率是95.1％。
In this paper we investigate the methods for recognizing tones in continuous speech based on syllableunit continuous HMMs. We proposed the new methods of feature extraction and tone recognition for Chinesecontinuous speech. The tone recognition was conducted for ten speakers. The average rate of 95.1% was obtained forthese speakers.
出处
《信号处理》
CSCD
2000年第1期20-23,共4页
Journal of Signal Processing
关键词
汉语连续语音
声调识别
基频
语音识别
Chinese continuous speech, Tone recognition, Fundamental frequency Continous HMM
