汉语耳语音孤立字识别研究

Isolated word recognition in Chinese whispered speech

耳语音识别有着广泛的应用前景,是一个全新的课题。但是由于耳语音本身的特点,如声级低、没有基频等,给耳语音识别研究带来了困难。本文根据耳语音信号发音模型,结合耳语音的声学特性,建立了一个汉语耳语音孤立字识别系统。由于耳语音信噪比低,必须对其进行语音增强处理, 同时在识别系统中应用声调信息提高了识别性能。实验结果说明了MFCC结合幅值包络可作为汉语耳语音自动识别的特征参数,在小字库内用HMM模型识别得出的识别率为90．4％。

The whispered speech recognition is a new subject which has wide applications. However, the characteristics of whispered speech such as its low sound pressure level and the lack of fundamental frequency bring difficulty to the whispered speech recognition. In this paper, a Chinese isolated word recognition system is established based on the source-filter generation model combined with the acoustic characteristics of whispered speech. In addition, the speech enhancement algorithm is added to the system to improve the SNR of whispered speech, and the tone information is implemented to acquire better recognition performance. The experimental results demonstrate that the MFCC combined with the amplitude contour features can be used as efficient parameters for the Chinese whispered speech recognition. A recognition rate of 90.4% is obtained when a small Chinese isolated word database is tested using HMM approach.