语音增强中小波收缩参数选择分析

Wavelet Shrinkage Parameters Selection in Speech Enhancement

小波收缩用于语音增强时,收缩参数直接关系到增强效果。为了在小波语音增强中得到可靠的参数选择方案和选择原则,本文引入正交试验方法,并通过正交试验结果修正方法得到可分析的数据结果,再根据数据结果进行最优参数组合和参数重要性分析。分析发现,阈值算法和阈值函数对语音增强的结果有重大影响,且重要程度受到浊音/清音等语音性质的影响;SURE阈值算法和软阈值函数是较好选择。本文还得到了其他小波增强参数的相关结论。

Speech enhancement is directly influenced by wavelet parameters selected in the process of wavelet shrinkage. However, hundreds of combinations generated by five wavelet shrinkage parameters must be selected. In order to obtain reliable parameter selections and principles for selecting, the orthogonal experiment and the result correction in parameter selecting are used for importance order analysis. It is found that the threshold algorithm and the function are very important for enhancement effects, and their degrees of the importance vary with characteristics of the voiced speech or the unvoiced speech. The analysis shows that SURE threshold algorithm and the soft threshold function are better than other considered parameters. Additionally, conclusions about other wavelet shrinkage parameters prove that all these results are meaningful to future wavelet shrinkage implementations