On the combination of redundant and zero-redundant channel error detection in CELP speech-coding

In this paper a new algorithm is described for selective detection and handling of speech codec parameters which were corrupted by bit-errors on the transmission channel. A combination of classical forward error detection schemes using additional (redundant) bits and parameter-correlation based techniques without redundant bits (zero-redundant error detections) is used for this purpose. The algorithm is optimized by informal listening tests rather than by maximization of mathematically tractable measures (e.g. SNR) which usually do not reflect subjective speech quality well. No additional delay and almost no additional memory and complexity is required for the new algorithm. The speech quality resulting at the decoder output is strongly improved compared to standard bad-frame handling techniques if coded speech is transmitted over disturbed channels, e.g. the GSM-full-rate channel which is used for performance evaluation.