Low bit and variable rate speech coding using local cosine transform
Dong Enqing, Zhao Heming, Yongli Li · 2004
An average1.6kb/s low bit and variable rate speech coder based on local cosine transform (LCT) algorithm for a two-way conversational speech is designed for the first time in the paper. The result of the voice activity detector (VAD) based on support vector machine (SVM) and the classification method of the voicing modes of the GSM half rate standard for active speech are adopted in the design of the variable bit rate coder. The moderately voiced mode and the strongly voiced mode of the voicing modes are combined as a voicing mode, the new combined voicing mode is named as a moderately and strongly voiced mode. A few segment vector quantizers of the LCT coefficients for each voicing mode and silence voicing frame (background noise) are employed, and LGB algorithm is applied to design the codebooks. A tree fast search technique is used to select the vector of the LCT coefficients for each segment. The evaluation using subject informal listening tests and a few object parameters indicates that the speech quality (intelligibility and naturalness) of the designed speech coder is better than that of the FS1015 standard coder. The new coder has higher robust than the FS1015 standard coder, which is suitable for speech coding in any environments.