Amplitude modulated sinusoidal models for audio modeling and coding

Abstract

In this paper a new perspective on modeling of transient phenomena in the context of sinusoidal audio modeling and coding is presented. In our approach the task of finding time-varying amplitudes for sinusoidal models is viewed as an AM demodulation problem. A general perfect reconstruction framework for amplitude modulated sinusoids is introduced and model reductions lead to a model for audio compression. Demodulation methods are considered for estimation of the time-varying amplitudes, and inherent constraints and limitations are discussed. Finally, some applications are considered and discussed and the concepts are demonstrated to improve sinusoidal modeling of audio and speech.