Minimizing state error propagation in low-bit rate speech codec for voice over IP

Today's Internet and intranets are packet-switched networks where the Internet Protocol (IP) is the most widespread of all network protocols in use. Such network protocol poses serious constraints on the real-time transmission of packets such as in the case of voice applications, namely Voice over IP (VoIP) applications. The "best effort" delivery mechanism and the lack of guarantee of Quality of Service (QoS) of packet networks are known to cause packet arrival problems that result in packet losses. Obviously, the loss of packets impairs the quality of the speech at the receiving end. In this thesis, we focus on two main areas in the implementation of VoIP systems. Initially, we review speech compression techniques to better understand the operation and characteristics of speech codecs. This allows us to select a promising speech codec, namely, the ITU-T G.729A speech codec that will be used during our demonstrations and investigations. Secondly, we review the operation of packet-switched networks, more specifically IP networks, for the purpose of understanding the degradation effects they cause to VoIP systems. With this knowledge, we formalize a set of constraints and requirements that allows us to properly analyze the effect of packet losses over IP networks. Finally, we propose a closed-loop over the network method to assist the codec in improving its speech quality performance in periods of packet losses. The method, named State Error Correction (SEC), is described in detail and its performance is assessed through simulations