First International Conference on Spoken Language Processing (ICSLP 90)

Kobe, Japan
November 18-22, 1990

Constrained-Stochastic Excitation Coding of Speech at 4.8 Kb/s

Yair Shoham

Signal Processing Research Department, AT&T Bell Laboratories, Murray Hill, NJ, USA

This paper proposes a method for enhancing the performance of Codebook-Excited Linear Predictive (CELP) coders. It is based on the observation that the codebook-driven excitation in these coders is noisy and that the noisy component is not adequately filtered by the LPC filter. It is proposed to adaptively constrain the amount of the noisy excitation by linking its level to a performance index of the long-term (pitch-loop) sub-system. This operation reduces the noisy effects of the excitation, enhances the synthesized speech periodicity and hence, the perceptual quality of the coder. Listening test results are presented to demonstrate the subjective improvement of this coder over the basic CELP. The CSEC technique has been implemented in various AT&T coders at 4.8 to 8.0 Kbps, including low-delay CELP, with both stochastic and trained codebooks. Noticeable improvement in speech quality has been achieved. The technique has also been incorporated in the proposed federal standard PFS1016 4.8 Kbps coder.

Full Paper

Bibliographic reference.  Shoham, Yair (1990): "Constrained-stochastic excitation coding of speech at 4.8 kb/s", In ICSLP-1990, 645-648.