基于MBE模型的基音周期估计算法改进

Improved on Pitch Period Calculation Based on MBE Model

多带激励(multiband excitation, MBE)模型是低速率语音编码中最为成功的模型之一,基音周期是MBE模型中的重要参数.基于MBE模型,首先研究了语音信号的基音周期估计算法,包括语音信号预处理、基音周期粗搜索与细搜索;然后,在此基础上,针对基音粗搜索中可能出现的基音周期加倍或减倍现象及细搜索过程中运算量过大等问题,提出了相应的动态规划平滑和细搜索改进算法,从而使基于MBE模型的基音估计算法更具实用性.

MBE model is one of the most successful models in low bit rate speech coding, and pitch period is one of the most important parameters in MBE model. Based on the model of MBE, the paper researches the arithmetic of pitch period calculation, which includes speech signal preprocessing, initial pitch estimation, pitch refining. Then on this basis, aiming at the problems in the initial pitch estimation such as wrong pitches in initial pitch estimation, the paper puts forward dynamic programming smoothing algorithm and improves refining arithmetic, so the pitch estimation of MBE model will be more useful.