Voice over IP and Vocal Recognition

This paper deals with the effects of packet loss and voice data compression on vocal recognition over IP connections. One contribution of this paper consists in diagnosing most precisely the problems due to the compression and the packet losses for two different recognition tasks: the automatic speech recognition and speaker recognition. Another contribution corresponds to the proposal for recovering techniques in order to improve the robustness of systems under significant packet losses conditions. From the diagnosis, a more important degradation due to the compression on the speaker verification task was noted. The experimental results show that the techniques of interleaving based on the transmitter combined with the interpolation based on the receiver, prove to be the most efficient