Packet loss protection for interactive speech object rendering: A multiple description approach

Abstract

This paper investigates the application of novel packet loss protection schemes to compress mixtures of speech sources for interactive real-time audio services such as spatial teleconferencing. Hybrid Forward Error Correction (FEC) and Multiple Description Coding (MDC) packet loss protection techniques are applied to the outputs of a psychoacoustic-based Analysis-By-Synthesis (PABS) coder designed for such applications. The protection approaches split the coder outputs into two descriptions that are separately protected using the hybrid FEC-MDC techniques. Perceptual Evaluation of Speech Quality (PESQ) measurements compare the performance of different protection schemes for a range of typical packet loss conditions. Results indicate the proposed scheme maintains the perceptual quality of the speech sources across a wide variety of packet loss conditions.