基于改进二次谱基音检测的时长规整算法

A Time Scale Modification Algorithm for Speech Based on Pitch Detection in Frequency Domain

为了降低经典自相关语音时长规整算法的计算复杂度,分析了自相关时长规整算法的计算复杂原因,通过改进二次谱基音检测方法取代经典自相关方法中的基音同步自相关搜索运算,提出了一种新的基于改进二次谱基音检测的语音时长规整算法,并给出了该算法的详细步骤。通过波形比较以及MOS打分实验表明,与经典自相关语音时长规整算法相比,该算法在保证时长规整后语音质量的前提下,有效降低了整体计算复杂度。

For reducing the computation of SOLAFS for Time Scale Modification of Speech,the self-correlation operation in SOLAFS is replaced by a pitch detection method based on power spectrum reprocessing modified.In this paper,a novel method for TSM based on pitch power spectrum reprocessing is proposed consequently,which provides superior speech quality and reduces the computation complexity,compared with SOLAFS.