论文部分内容阅读
为了解决高质量散布脉冲码激励线性预测语音编码算法中的代数码书搜索复杂度高的问题,通过在候选的脉冲位置上保持目标信号和脉冲响应的相对最大相关,提出了一种不降低主观话音质量的多路径散布脉冲搜索算法,计算负载从19.66 MOPS减少到0.069 4:MOPS,大幅度减轻了散布脉冲码书的计算负载.主观和客观听力测试表明,算法的改进前后的合成语音质量没有感性差别.
In order to solve the problem of high searching complexity of the algebraic codebook in the linear predictive speech coding algorithm of high-quality scattered pulse code excitation, a method is proposed to maintain the relative maximum correlation between the target signal and impulse response at the candidate pulse positions. The multipath spreading pulse search algorithm for voice quality reduced the computational load from 19.66 MOPS to 0.069 4: MOPS, which greatly reduced the computational burden of spreading the impulse codebook.The subjective and objective hearing tests showed that the synthesized speech quality before and after the algorithm improvement Sensitive differences.