A Concealment Technique For Missing VoIP Packets Across Non-deterministic IP Networks

Abstract

Voice over Internet Protocol (VoIP) transmits packetized voice frames over a packet-switched (PS) network. In practice, data units may be delayed when a network carries more traffic than it can handle, leaving gaps within the audio segment. In this scenario, the receiving terminal cannot request the lost packets to be retransmitted since VoIP communication is real-time. For such gaps, except when a concealment technique is applied, an indeterminate silence period may occur, giving an impression that the call has been terminated. The usual technique for addressing this problem is to synthetically
regenerate the lost packet by estimating the waveform of the original signal such that the person who listens does not notice the noise. Earlier works based on pitch waveform replication, pattern matching, and phase matching reported good performances. However, their perceived regenerated signal quality deteriorates severely with an increased duration of packet loss, reflecting a lack of continuity between the reconstructed and the later packets. This paper presents an
advanced receiver-oriented algorithm for missing VoIP packet concealment, based on Time-Scale Modification (TSM), derived from Waveform Similarity Overlap Add (WSOLA) to restore the missing packets. Leveraging on the WSOLA, the Variable Analysis Segment Waveform Similarity Overlap Add (VASWSOLA) was evolved. The VASWSOLA is speech
characteristic dependent, particularly to the pitch of a specific person who speaks. Therefore, a pitch estimation is implemented before the approximation of an analysis segment size, Sa. Finally, subjective assessment tests by a group of
listeners established that the proposed technique remarkably enhances the quality of the regenerated speech.