小波匹配追踪的语音信号时频建模

摘要

Sinusoidal model based on wavelet multi-resolution analysis can not get better modeling of speech signal with low-sinusoidal context. It proposes the transient modeling of speech signal by matching pursuit with a wavelet-based dictionary. After building wavelet-based dictionary, diversity mutation genetic matching pursuit is used to find the best wavelet atom, and speech signal is reconstructed by a series of the best wavelet atoms. Comparative analysis between the two models shows that residual of the proposed model is significantly reduced.%小波多分辨率的正弦模型对正弦成分含量较低的语音信号不能较好建模,提出小波匹配追踪的瞬时建模方法.建立小波原子词典,利用多样性变异算子匹配追踪寻找最佳匹配小波,最后将语音信号表示成一系列匹配最佳小波的累加.仿真实验针对正弦成分含量较低的语音信号对两种模型作比较,结果显示提出模型的信号重构残差值较小波变换的正弦模型有明显降低.