Third International Workshop on Models and Analysis of Vocal Emissions for Biomedical Applications (MAVEBA 2003)

Florence, Italy
December 10-12, 2003

On Packet Loss Concealment Using Time-Varying Speech Analysis

Keiichi Funaki

Computing and Networking Center, University of the Ryukyus, Nishihara, Nakagami, Okinawa, Japan

Demand of IP telephone is increasing more and more as broadband IP network is being commonly used by many people. In VoIP, packet loss concealment (PLC) is one of the key subjects to keep speech quality since packet loss occurs in IP network. PLC methods based on Linear Predictive coding (LP) method have been proposed in which LP coefficients and LP residual are repeated to recover speech corresponding to the packet loss. However the repetition would not perform well in any speech frames. This paper presents a novel PLC method based on time-varying speech analysis and synthesis, in which AR parameters can be predicted owing to its time-domain function of AR parameter. Three kinds of AR parameter prediction methods by means of time-varying analysis are evaluated subjectively and objectively and novel PLC scheme switching the AR prediction methods with respect to F0 prediction gain is proposed.

Index Terms. VoIP, PLC, Time-varying speech analysis

Full Paper (reprinted with permission from Firenze University Press)

Bibliographic reference.  Funaki, Keiichi (2003): "On packet loss concealment using time-varying speech analysis", In MAVEBA-2003, 79-82.