An error correction approach which is based on a hidden Markov model (HMM) is proposed. The occurrence probability of a code sequence, which is delivered by the HMMs, is used as the measure for the maximum a posteriori probability (MAP) algorithm. The MAP algorithm is based on the assumption that the source is a discrete-time finite-state Markov process, and the HMM which models a Markov source is well suited for speech data. Therefore this combination would be useful for a speech coding system. The proposed approach is adapted to the code sequence quantized line spectrum frequency (LSF) parameters. When the code sequence is sent over a binary symmetry channel (BSC), the proposed approach with 16-state HMMs improves in code error rate and degradation of the cepstrum distortion by about 27% and 39% respectively for 3% random errors.
[1]
Robert M. Gray,et al.
Joint source and noisy channel trellis encoding
,
1981,
IEEE Trans. Inf. Theory.
[2]
Javier Garcia-Frías,et al.
Combining hidden Markov source models and parallel concatenated codes
,
1997,
IEEE Communications Letters.
[3]
Michael W. Marcellin,et al.
Trellis-coded vector quantization
,
1991,
IEEE Trans. Inf. Theory.
[4]
T.E. Fuja,et al.
Channel codes that exploit the residual redundancy in CELP-encoded speech
,
1996,
IEEE Trans. Speech Audio Process..
[5]
John H. L. Hansen,et al.
Discrete-Time Processing of Speech Signals
,
1993
.