On Using Digital Speech Processing Techniques for Synchronization in Heterogeneous Teleconferencing

As the popularity of multi-functional communication devices grows, traditional audio conferencing now may involve heterogeneous teleconferencing devices, including POTS phone, VoIP phones, dualmode smart phones, and so on. During a multi-party audio conference involving heterogeneous devices, it is possible that a video conference is held concurrently involving a subset of devices capable of processing video streams for better the conferencing experience. In such a scenario, the need for synchronization between circuit-switched audio streams and packet-switched video streams arises. While the problem of audio-video synchronization has been extensively investigated in related work, existing solutions are limited to synchronization in packet-data networks and hence are not applicable in the target environment. In this work, we consider the problem of supporting such an overlay video conference among dual-mode phones. We first transform the audio-video synchronization problem into the problem of synchronizing circuit-switched and packet-switched audio streams. We then propose an end-to-end solution for audio synchronization that is transparent to the heterogeneous network protocol suites involved. We investigate synchronization algorithms based on digital speech processing using different acoustic features of the speech signal in the waveform, cepstrum, and spectrum domains. We evaluate the effectiveness of different algorithms under various impairments including codec distortion, line noises, packet losses, and overlapping utterances. Evaluation results show a promising direction for using DSP-based algorithms to address the synchronization problem across heterogeneous telephony systems.