基于小波变换的重叠语音基频提取及声调识别

Overlapping speech pitch extraction and tone recognition based on wavelet transform

提出一种基于小波变换的重叠语音基频提取及声调识别的方法。利用小波的伸缩和时移特性，通过对重叠语音做多尺度的小波变换，可以有效地提取重叠语音中各自的基音频率，并在此基础上实现声调的识别。实验表明，此方法是有效的，是重叠语音基频提取及声调识别的一种新途径。

Based on wavelet transform, this paper submits a new approach to overlapping speech pitch extraction and tone recognition. Taking advantage of retractility and time-delay characteristics of wavelet, it can detect each pitch in overlapping speech effectively and implement tone recognition by multiscale wavelet transform on overlapping speech. Results of experiment showed the approach is efficient.