Robust decoding of speech line spectral frequencies over packet networks
Paul Rondeau · Mspace (University of Manitoba) · 2007
The problem of transmitting line spectral frequencies (LSF) generated by the Federal Standard 1016 CELP speech encoder over a packetloss net-wo¡k is considered.Multiple description (MD) coding techniques are investigated, which reduce the spectral distortion (SD) ìn speech due to packet losses while using the same transmission rate as the standard CELP encoder.We focus on exploiting tl-re residual redundar-rcy of the encoder ouþut to estimate lost packets at the receiver, by using hidden Markov modeling of the encocler ouþut and estimation based on the forward-backward algorithm.In particular, the problem of optimizing index assignments for Markov decoders is addressed.Experimental results are presented which compare the proposed techniques with other known techniques, such as linear estìmation and Gaussian mixfure modeling.It is demonstrated that the proposed Markov technique averages 2.18 dB of SD when one description is lost, compared to interpolation of odd-even split LSFs at 2.99 dB.