Implementation of advanced audio coder on DSP processor

Abstract

During the last three years an obscure signal compression technology known to the technical world as MPEG layer 3 has become a household term: mp3. mp3 and its sister algorithms, which attempt to preserve the fidelity of the original music recording while carrying that signal in fewer bits, have made it practical to distribute music recordings over the network, and in doing so have called into question long-held views of ownership of intellectual property. In this overview talk I will discuss the principles underlying most modern audio coding algorithms, as exemplified by the MPEG Advanced Audio Coder algorithm, the successor to mp3 (at least in the technical world). I will compare AAC performance with the performance of other audio algorithms, and will demonstrate audio coding at various bit rates using AT&T's AAC implementation. In the second half of the talk, I will discuss the effect of packet loss on compressed audio data streams, and demonstrate the effectiveness of simple strategies for concealing packet loss. Using this as a starting point, I will discuss an IETF draft which uses error concealment as a form of bandwidth control. Time permitting, I will demonstrate Verdi, a multimedia streaming system developed at AT&T Labs Research, which uses AAC as its preferred audio compression technology.