An Inter-frame Correlation Based Error Concealment of Immittance Spectral Coefficients for Mobile Speech and Audio Codecs

This paper proposes an inter-frame correlation based error concealment approach for hybrid CELP (Code Excited Linear Prediction) and transform codec's with both good speech and audio quality at moderate bit rates. The proposed scheme is designed to overcome the main challenge due to the diversified characteristics of input signals. The underlying idea is to employ the inter-frame correlation of previous neighborhood frames to circumvent the pitfalls of referring to the unrelated frames, and to enable effective prediction of ISF (Immittance Spectral Frequencies) spectrum coefficients of missing frames from the immediate relative history using linear regression approach. Objective and subjective evaluation results for the proposed approach, in comparison with existing technique of AVS-P10 (Audio Video coding of China Standard Part 10 -- Mobile Speech and Audio Codec), provide strong evidence for gains across a variety of speech and audio signals.