ADAPTIVE SPEECH QUALITY IN VOICE-OVER-IP COMMUNICATIONS
The quality of VoIP communication relies significantly on the network that transports the voice packets because this network does not usually guarantee the available bandwidth, delay, and loss that are critical for real-time voice traffic. The solution proposed here is to manage the voice-over-IP stream dynamically, changing parameters as needed to assure quality. The main objective of this dissertation is to develop an adaptive speech encoding system that can be applied to conventional (telephony-grade) and wideband voice communications. This comprehensive study includes the investigation and development of three key components of the system. First, to manage VoIP quality dynamically, a tool is needed to measure real-time changes in quality. The E-model, which exists for narrowband communication, is extended to a single computational technique that measures speech quality for narrowband and wideband VoIP codecs. This part of the dissertation also develops important theoretical work in the area of wideband telephony. The second system component is a variable speech-encoding algorithm. Although VoIP performance is affected by multiple codecs and network-based factors, only three factors can be managed dynamically: voice payload size, speech compression and jitter buffer management. Using an existing adaptive jitter-buffer algorithm, voice packet-size and compression variation are studied as they affect speech quality under different network conditions. This study explains the relationships among multiple parameters as they affect speech transmission and its resulting quality. Then, based on these two components, the third system component is a novel adaptive-rate control algorithm that establishes the interaction between a VoIP sender and receiver, and manages voice quality in real-time. Simulations demonstrate that the system provides better average voice quality than traditional VoIP.
Advisor:Dr. Stephen Walters; Dr. Martin Weiss; Dr. David Tipper; Dr. Joseph Kabara; Dr. Richard Thompson (Advisor)
School:University of Pittsburgh
School Location:USA - Pennsylvania
Source Type:Master's Thesis
Date of Publication:05/14/2008