Detection of synchronisation errors for the G.728 decoder

A new method for the fast detection of frame synchronisation errors inside the G.728 speech decoder is described. The proposed algorithm uses a priori knowledge about internal decoder parameters. The G.728 frame synchronisation fails, if an octet slip in the transmission system occurs. This error type causes a shift of the word boundaries in the G.728 data stream. The decoder produces in this case a loud and noisy output signal. The proposed algorithm can be implemented as an add-on in a regular G.728 decoder, without changing the standard operation of the codec. Three classification methods, the Euclidian distance classifier, McCulloch-Pitts network and feedforward neural network are compared with respect to their performance and complexity.