基于自适应加权谱内插的宽带语音编码算法

Wideband Speech Coding Algorithm Based on Adaptive Interpolation of Weighted Spectrum

提出了一种基于自适应加权谱内插 ( STRAIGHT)的宽带语音编码算法。输入的语音信号首先经过STRAIGHT分析得到精确的基频参数和谱参数 ,然后通过时域抽取和频域建模实现有效的编码压缩。在时域抽取时采用的区别于传统编码算法固定帧长的自适应可变帧长方法 ,使得编码存储量可以根据实际语音变化情况得到更加合理的分配。主观测听结果表明 ,该算法针对 1 6k Hz采样的语音信号 ,在 6kbps码率上可以取得与AMR-WB( G.72 2 .2 )在 8.85 kbps时的相当的音质效果。此外 ,该算法还具有对恢复语音的时长。

Based on speech transformation and representation using adaptive interpolation of weighted spectrum (STRAIGHT), a wideband speech coding algorithm is presented. The input speech signals are firstly decomposed into pitch parameters and spectral parameters by STRAIGHT,and then compressed effectively by sampling in temporal domain and modeling in frequency domain. Because of the introduction of adaptive sampling with variable frame lengths, the bitrates can be more reasonably allocated accoding to the actural movement of speech signals. Subjective listening test demonstrates that the decoded quality of proposed algorithm at 6 kbps for 16 kHz sampled speech signal corresponds to that of AMR-WB(G.722.2) at 8.85 kbps. Besides, the method has flexible modification ability on duration, pitch and spectrum of decoded speech. So it can be widely applied in the fields, such as speech synthesis with parametric modification, voice conversion and so on.