基于小波变换和参数滤波的音素分段算法

论文首先分析了小波的时频特性,基于该特性对语音信号进行小波域滤波,提出对听觉感知有效的频率分量,然后用参数滤波方法进行分段。参数滤波的基本思想是以一个变化的参数对信号进行滤波,得到信号在不同频带中的分量。可以证明若滤波参数以一定的规律变化,则这些滤波分量的一阶自相关表示了信号的相关结构。实验表明对上述经小波域滤波后的频率分量进行基于参数滤波的音素分段会得到较准确的分段效果。

A New Speech Segmentation Algorithm Based on Wavelet Transform and Parametric Filter

At first,we study the time-frequency property of the wavelet transform.Based on this property,we filter the speech signal in wavelet domain.Its objective is to abstract the signal components that are important in hearing.Then we use the parametric filter(PF) method to segment.The PF method is motivated by the fact the correlation structure of astationary signal can be characterized by the signature of certain output statistics from a designed filter bank.It is proved in our experiment that when filtering the raw speech signal firstly,we can get a more accurate segment result.