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

Abstract

This paper presents a new framework for compression and transmission of simultaneously occurring audio objects through packet loss channels in order to maintain user Quality of Experience (QoE). The audio objects are compressed into two mono mixtures, by exploiting sparsity of multichannel audio signals to identify the two most dominant time-frequency components in terms of an energy measure. These mixtures are further compressed using the MP3 audio codec, with the optimised transmission model selected from several channel coding models based on the Forward Error Correction (FEC) and Multiple Description Coding (MDC) packet loss protection techniques. Audio objects can be recovered robustly from any received description(s) allowing real-time selective reproduction at the listeners' end. Results from spectral distortion measurements indicate the proposed scheme maintains the perceptual quality of the audio objects across a wide variety of packet loss conditions.