在多脉冲线性预测编码的基础上,本文提出了位置无关脉冲搜索算法。该算法不需要搜索脉冲位置,而是根据给定的脉冲位置一次性解出脉冲幅度矢量。这就保证了得到的脉冲组合在最小二乘意义下是最优的,为改进合成语音质量提供了理论基础。进而在激励脉冲与位置无关的理论基础上,提出了定点脉冲线性预测编码方法。对所提出的算法在MATLAB下进行了仿真,仿真结果发现位置无关脉冲搜索算法得到的合成语音质量优于序贯法,编码时间也要比序贯法短。定点脉冲线性预测编码方法可以在2.7 kbps的编码速率下获得与G.729相近的合成语音。
In this paper, a location-independent pulse search algorithm based on multi-pulse linear prediction coding (MP-LPC) is presented. This algorithm can solve all the amplitudes of the pulses at a time according to given pulse locations which need not be solved using analysis-by-synthesis procedure. This can ensure that the solved pulses are optimal in a least square sense, which provides the theoretical foundation for improving the quality of synthesized speech. And then based on the location-independent pulse search algorithm, fixed- location pulse linear prediction coding (FLP-LPC) method is presented. The presented algorithm is simulated in MATLAB, the results show that the synthesized speech using the pulses solved by location-independent pulse search algorithm has better speech quality and shorter coding time than the sequential method. FLP-LPC can obtain the synthesized speech closed to C.729 at the coding rate of 2.7 kbps.