Isolated word recognition in Chinese whispered speech
Yang Li · Applied Acoustics · 2006
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.