论文部分内容阅读
提出了一种基于压缩感知的考虑语音帧间信息的语音转换算法。根据连续多帧语音的线谱对参数所构成的矢量在离散余弦变换域具有稀疏性,利用压缩感知技术对该矢量压缩成短矢量,并将该压缩后的短矢量作为特征参数训练语音转换函数。实验测试结果表明,选择合适的语音帧数时,该算法的性能要比传统的采用加权频率卷绕的转换算法提高3.21%。这说明,充分有效地利用语音帧间的相关信息会使转换语音保持更稳定的帧间声学特性,有利于提高语音转换系统的性能,
A speech conversion algorithm based on compressive perception considering speech inter-frame information is proposed. According to the spectrum of continuous multi-frame speech, the vector composed of the parameters is sparse in the discrete cosine transform domain, the vector is compressed into a short vector by using compressed sensing technique, and the compressed short vector is used as the feature parameter to train the speech transfer function . Experimental results show that the performance of this algorithm is 3.21% higher than that of the traditional algorithm using weighted frequency winding when selecting the appropriate number of speech frames. This shows that full and effective use of the relevant information between the speech frames will make the converted speech to maintain a more stable inter-frame acoustic characteristics, which will help to improve the performance of the speech conversion system,