Sinusoidal Speech Modelling for Packet Based Transmission

The scope of this research project is packet-based transmission of speech over IP networks, i.e. Voice over IP (VoIP). This application gives rise to some special requirements to the speech coding. Besides obtaining a low bit rate and a high speech quality, the possibility of packet loss suggests that packets should be decoded independently, i.e. without needing information from prior or posterior packets. This motivates the use of sinusoidal speech coding, since this method can provide a high speech quality at a low bit rate without using inter-frame information. A major challenge in VoIP is compensating for lost and delayed packets. For the purpose of dealing with delayed packets the receiving end stores recently arrived packets in a jitter buffer before play-out. This approach solves the delayed packet problem, as long as the delay is not longer than the length of the jitter buffer. When the delay is greater than the jitter buffer length, however, the packet is considered lost, and must be compensated for. Usually, this is attempted by estimating the missing packet from the prior and posterior packets. However, when using the sinusoidal model a promising alternative is to stretch the surrounding packets in time into the missing packet interval. A more advanced scheme is to stretch preceding packets to allow more time waiting for the missing packet [C.A. Rødbro, M.G. Christensen, S.V. Andersen, S H. Jensen 2003]. (Christoffer Asgaard Rødbro, Søren Holdt Jensen, Søren Vang Andersen)