Multimode Tree Coding of speech with backward pitch prediction and perceptual pre- and post-weighting

Ying-Yi Li, Jerry D. Gibson · 2012

A low delay and low complexity Multimode Tree Coder with backward pitch predictor is proposed. For the Multimode Tree Coding, the speech is classified into five different modes, and each mode is coded at a suitable bit-rate using a tree coder with perceptual pre- and post-weighting filters. In order to improve the speech quality without increasing the delay, a backward pitch predictor is added into the the Voiced mode. The results show that the pitch predictor does improve the PESQ-MOS. In addition, the PESQ-MOS of the Multimode Tree Coder is equivalent to the PESQ-MOS of AMR-NB at 12.2 kbps and G.728 at 16 kbps while the computational complexity is lower than AMR-NB and G.728 and the delay is lower than AMR-NB.

Read the paper · More papers on PaperTik