摘要
在HMM算法的基础上引入了小波去噪理论,从而提高了原始语音的信噪比和最终识别率.由于分别对每段语音去噪并进行端点检测,大大降低了运算量,因而减少了训练时间,达到了较好的识别效果.通过与DTW算法的对比,证明了改进的HMM算法在非特定人语音识别中的良好效果.
The theory of wavelet denoising is applied in the hidden Markov model （HMM） algorithm, which enhances the signal-to-noise ratio （SNR） and recognition accuracy. By denoising and endpoints detecting in each speech, both computational load and training-time are reduced, and the recognition performance is improved. Compared with the dynamic time warping （DTW）, the improved HMM algorithm has a better recognition performance in speaker-independent speech recognition.
出处
《四川大学学报（自然科学版）》
CAS
CSCD
北大核心
2007年第1期69-72,共4页
Journal of Sichuan University(Natural Science Edition)
关键词
HMM
非特定人语音识别
小波去噪
HMM, speaker-independent speech recognition, wavelet denoise
作者简介
李锦（1982-），男，2004级硕士研究生，研究方向为数字信息处理。
