CELP coding at 4.0 kb/sec and below: improvements to FS-1016

While code excited linear prediction (CELP) speech coders can produce high levels of output speech quality at rates near 4 kb/s, they may not be suitable for toll quality communications. The authors propose several modifications to improve the quality of the standard CELP algorithm while simultaneously reducing its transmission rate. The modifications include a multimode excitation that increases the convergence rate of the adaptive (pitch) codebook. A low-complexity spectral vector quantization algorithm is also developed that reduces the coding rate and decreases the spectral distortion. SNR improvements of 2 dB and a significant reduction in perceptual artifacts have been observed.< >