基于线性预测残差倒谱的多语音基音频率检测

A Multi-pitch Detection Algorithm Based on Cepstrum of Linear Prediction Residual

本文提出了一种基于线性预测残差倒谱的多语音基音频率检测算法,该算法首先对混合语音信号进行线性预测分析,进而计算预测信号与原混合信号的残差,并对残差信号做倒谱变换,得到混合语音信号的线性预测残差倒谱;然后在该信号的残差倒谱中,结合图像处理的技术,利用语音信号基音倒频匹配法检测出多语音信号的基音频率;最后在基音标定的过程中,本文算法利用语音信号的连续特性,依据信号基音频率前后差距变化最小原则标记出各基音所属话者。实验结果表明,本文提出的算法在弱回声及无回声的情况下能快速有效地从单声道混合语音信号中检测出多语音基音信息。

An algorithm based on the cepstrum of linear prediction （LP） residual for pitch detection is presented in this paper. The cepstrum of linear prediction residual of the speech signal is used to be the information for pitch determination. Specifically, we firstly compute the linear prediction residual of the speech signals and the cepstrum of these residual. Then the multi-pitch information is detected by pitch matching from the cepstrum of linear prediction residual. Lastly we determined the owner of each pitch based on the principle of smallest change in frequency. The experimental results indicate that our method can effectively detect muti-pitch information from one microphone multi-speech signals of different speakers under the environment of anechoic or poor echo.