An analysis-by-synthesis linear predictive model for narrowband speech coding
F.F. Tseng · International Conference on Acoustics, Speech, and Signal Processing · 2002
An analysis-by-synthesis linear predictive model for narrowband speech coding is proposed. By introducing a perceptually weighted distortion measure between the original speech and the reconstructed speech, both the voiced/unvoiced decision and the pitch estimation/tracking are implicitly optimized. The excitation model includes two codebooks. The unvoiced codebook contains a collection of random sequences, and the voiced codebook contains a pitch synthesizer with a different pitch period and pitch gain. A 2.4-kb/s and a 1.2-kb/s speech coder are used as examples. An efficient method for closed-loop spectrum filter analysis is presented.>