Voice is the method of choice for real time communications. Voice is so important to human communications that we have constructed entire networks centered around voice, namely, the public switched telephone network (PSTN) and the analog/digital cellular networks. With the emergence of voice over the Internet Protocol (VoIP) and voice over wireless local area networks (Wi-Fi), a future voice call might consist of a digital cellular user communicating through a VoIP backbone network with a voice over Wi-Fi user, resulting in the connection of a digital cellular network through a VoIP network into a Wi-Fi link. Although the digital cellular networks utilize relatively sophisticated error detection and correction, unequal error protection, and error concealment, the error control, packet loss mechanisms, and packet loss concealment methods for VoIP and voice over Wi-Fi are relatively primitive. Furthermore, since each of these networks has different channel behaviors and may use different voice codecs and different network protocols, acceptable, bandwidth efficient voice service may not be achievable over such a connection. This research involves voice communications over tandem heterogeneous networks and voice over Wi-Fi links. We examine (1) improved voice codec tandem performance, by developing speech-mode-adaptive postfilters and perceptual distortion measures, by removing the postfilter and inserting additional filtering within the analysis-by-synthesis loop, and by shaping the excitation within the analysis-by-synthesis loop; and (2) improved packet loss concealment and reduced latency, by jointly designing packet loss concealment methods and protocols for wireless LANs and by developing packet loss smoothing algorithms to use "late" packets to improve later reconstruction. The evaluation of the end-to-end network performance for voice communications uses the NS2 network simulator and voice quality metrics.
