一种基于小波变换的混叠语音基音提取方法

小波变换对单一信号源的基音提取非常有效，但在多音源混叠语音的基音提取中则无法有效地获得基音信息。为解决这一问题，先从混叠语音信号中确定信号的个数以及每路信号的时间间隔范围，之后再对其进行奇异点的分离，以便提取各源信号的谐波分量。为简化计算并避免由于幅值的不同而影响搜索效果，对各次谐波的幅值进行了归一化、卷积等处理，最后使用小波变换精确地提取出混叠语音的基音。实验表明了这一方法的有效性。

Pitch Detection of Overlapping Speech Signals Based on Wavelet Transformation

Wavelet transform is very effective for picking up the pitch detection from a single source,but there is no effective way to obtain the pitch information from the pitch detection in overlapping speech with multi-sources.In order to solve this problem,the number of signals and the interval range of time-span of each signal source should be first determined from the overlapping speech signals,and then the separating singularities should be conducted in such a way as to pick up the harmonic components of all signals. To simplify the computation and avoid affecting search results caused by different amplitudes,the unified means and deconvolution are used to process all harmonics' amplitudes.Finally,wavelet transform is used to pick up the pitch of overlapping speech signals accurately.The experiment proves that this method is valid.