语音编码中的递归神经网络技术研究  

Study Of Recurrent Neural Networks Technic in speech coding

作　　者：朱文育[1] 王炳锡[1] 

机构地区：[1]郑州解放军信息工程大学

出　　处：《微计算机信息》2005年第11X期203-204,26,共3页Control & Automation

基　　金：国家自然科学基金资助项目;编号:60372038

摘　　要：众所周知,线谱对(LSP-LinearSpectrumPair)系数是一种线性预测系数,它表征的是语音谱包络。在时域中它的谱插值性能良好,但是它的插值间隔一般都限制在20～30毫秒之间。为了解决这个问题,本文介绍一种使用递归神经神经网络(RNN-RecurrentNeuralNetworks)来对线谱对系数进行插值的算法。实验结果表明,使用递归神经网络可以使插值的间隔增加到100毫秒而不明显降低合成语音的质量。It is well known that the LSP coefficients which represent the speech spectrum envelope as one of the linear prediction coefficients,show a good performance of spectral interpo-lation along the time axis , but it is also known that the duration of the interpolation is limited up to 20-30ms. To resolve this problem,recurrent neural networks were applied to interpolate LSP coefficients.the experiment results show that this method can increase the duration of the interpolation to about lOOms without so much degradation of the synthesis speech quality.

关 键 词：线谱对系数 谱插值 递归神经网络 

分 类 号：TN912[电子电信—通信与信息系统]