SUMMARYThis paper proposes an error-correction algorithm for channel error using the hidden Markov model (HMM). The proposed method uses two probabilities calculated by the information source model (HMM) and the channel model, under the constraint of the received code sequence, and estimates the transmitted code sequence by maximum likelihood. Consequently, the robustness to error can be improved without adding error-correcting code by applying the proposed method to the preprocessing of the existing code system. Using the code sequence obtained by quantizing the LSP parameters, the reduction of cepstrum distortion in error correction by the proposed method is examined. It is found that the distortion can be reduced by approximately 39% by the proposed method (16-state HMM) for a code sequence with 3% random errors.
The relation between the perceptual noisiness of quantization error and the number of prediction coefficients was experimentally studied. It was revealed that a speech coder (a kind of APC) with ten coefficients shows considerably less noisiness than that one with four coefficients. Though improvement of the segmental SNR was only 1 dB, perceptual noisiness was improved by an equivalent value to a quantizing accuracy of 1.5 bits (approximately 9 dB). Taking this result into consideration, a coder scheme was proposed for special-purpose, narrow-band, wireless communcations. The narrow-band speech coder to be used in very noisy channels having a transmission error rate of 10−2 or 10−1 requires good intelligibility and less perceptual noisiness. It was confirmed that a source coder having eight or ten prediction coefficients and an error correction code for side information would meet the requirements. Even if the residual information fails to be transmitted, the source decoder is capable of synthesizing intelligible speech from only the side information including coefficients.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.