一种基于MBE的快速基音搜索算法研究

A Fast Pitch Detection Algorithm Based on MBE

多带激励声码器(MBE)是目前理想的语音编码方案之一,他的核心参数是基音周期,分析过程采用ABS方法,提高了语音参数提取的准确性,可在噪声环境中合成出自然度和可懂度很高的语音。针对其基音搜索算法计算量过大的缺点,首先提出了一种新颖的中心偏离函数计算法,该方法先对原始语音进行傅里叶变换得到原始语音谱,再针对各个基音候选值确定其相应于原始语音谱的子带划分,通过对比每个划分的中心偏离量,可以确定最佳基音周期;其次设计了一种简单有效的倍频消除算法,通过简单的循环计算,可以消除倍频现象;仿真结果表明,这两种方法的结合不仅大大降低了原算法的计算量,而且保持了他精度高的优点。

Multiband Excitation Vocoder(MBE) is one of the best coding at present,pitch is the most important parameter of MBE,pitch is obtained by the ABS process so that precision can be improved even in the noise environment,but the process of pitch dectection algorithm of MBE is too complex.In order to reduce the complexity of MBE pitch dectection algorithm in this article,we not only put forward a kind of novel algorithm named center variance function,but also propose a correction method to eliminate multi pitch.The speech signal will be transformed by FFT,then the transformed signal will be divided into some blocks according to the expected pitches,last,we will compare every center covariance of each expected pitch to get final pitch,the final pitch is corrected by a simple circle computation to correct multipitch,test indicates that the combination of the two methods mentioned above can greatly reduce the complexity of pitch detection algorithm of MBE,at the same time,the high quality of MBE saved.