The Influence of Clipping on the Performance of a Low Bit Rate Parametric Speech Coder

This paper presents a study of the influence of clipping on low bit rate parametric speech coder. By selecting three parameters of MELP vocoder: the fundamental frequency, voicing and LPC, the influence of different clipping degree on the MELP vocoder is studied. In the experimental stage, the performance of the MELP vocoder at different clipping degree are evaluated by using pitch estimation accuracy rate, voicing decision error rate and average Log-spectral distance. And then, the reasons which lead to a decrease in the performance of parameter extraction are analyzed. The experimental results show that clipping has a certain influence on the accuracy of pitch estimation and voicing decision. However, the accuracy of pitch estimation and voicing decision can keep high accuracy even when the clipping degree is large, which indicates that the pitch and voicing are strongly resistant to clipping. For LPC, the increase of the clipping degree increases the average Log-spectral distance, resulting in an inaccurate estimation of LPC.