利用一种新的谱扩展技术提高窄带语音质量

Quality Enhancement of Narrowband speech via a New Algorithm for Wideband Extension

提出了一种谱扩展算法,先对窄带语音进行线性预测编码(LPC)分析和多带激励(MBE)分析,得到窄带语音的线性谱频率(LSF)参数和清/浊音(U/V)判决信息;利用高斯混合模型(GMM)由窄带语音的LSF参数扩展得到高带语音的包络谱信息;对于激励信号,由窄带语音的U/V判决信息通过码本匹配得到高带的U/V判决,然后根据U/V判决信息由正弦模型合成出高带激励信号,并进而由LPC方法生成高带语音。最终的宽带语音由原始窄带语音和扩展得到的高带语音经子带合成得到。实验结果表明该算法扩展得到的宽带语音能有效提高语音信号的清晰度和可懂度。

A new algorithm of narrow-band speech spectrum extension is presented for quality en hancement.This algorithm uses LPC analysis to get the LSF parameters of the narrow-band(low-band)speech and MBE analysis to get U/V information.Spectrum envelope of high-band speech is obtained from the LSF parameters of narrow-band through GMM.U/V in formation of high-band is mapped from U/V decision of low-band.The high-band excitation sig nal is synthesized by sinusoidal model based on U/V information,and the high-band speech is synthesized by LPC model.Adding the high-band speech to the low-band,wide-band speech is synthesized.Experiments' results show that this algorithm can offer good performance wideband speech.[