Stochastic joint source-channel coding for low bit rate robust speech coding

Speech coders operating at low bit rates necessitate efficient encoding of the linear predictive coding (LPC) coefficients. In this paper, we propose a stochastic joint source-channel scheme developed for efficient and robust encoding of LPC coefficients in terms of LSF parameters. The encoding system, named LSF-SSCOVQ-RC, is an LSF encoding scheme based on a reduced complexity stochastic split vector quantizer optimized for noisy channel. For transmissions over noisy channel, we will show first that our LSF-SSCOVQ-RC encoder outperforms the conventional LSF encoder designed by the split vector quantizer (SVQ). After that, we applied the LSF-SSCOVQ-RC encoder (with weighted distance) for the robust encoding of LSF parameters of the MELP speech coder operating over a noisy channel. The simulation results will show that the proposed LSF encoder, incorporated in the MELP, ensure better performances than the original MELP MSVQ of 25 bits/frame; especially when the channel is highly disturbed.