基于Linux平台上线性预测语音编码器性能分析

The performance comparison of linear predictive vocoders on Linux platform

线性预测编码器是一类非常重要的语音编码器。文中主要描述了几种线性预测语音编码算法的软件及实现,即码激励线性预测编码(CELP)、低延迟CELP(LD-CELP)和混合励磁线性预测(MELP),编码速率分别为4.8、16、2.4kb/s。几种语音编码器的C语言程序已在Linux平台上可编译和执行,并对结果进行了以MOS为标准的主观测试。波形分析主要使用了Praat和Adobe Audition软件。结果表明,MELP和CELP的质量相当,而LD-CELP编码器的质量则要高得多,虽然是以牺牲较高的比特率为代价。

Linear predictive coders are a kind of significant vocoders.The speech coders using three linear predictive vocoding algorithms are described,i.e.CELP（code excited linear prediction）,LD-CELP（low delay-CELP）and MELP（mixed-excitation linear predictive）working at 4.8 kb/s,16 kb/s and 2.4 kb/s,respectively.The C programs for the three vocoders are complied and executed on the Linux platform and the results are subjectively tested with taking MOS as criterion.Waveform analysis mainly uses Praat and Adobe Audition software.The results show that the performances of MELP and CELP are of comparable quality,while the performance of LD-CELP is much higher,albeit at the expense of higher bit rates.