The Cancellation Principle in Acoustic Scene Analysis

Abstract

Cancellation is a useful “ingredient” to solve the problems of speech separation and acoustic scene analysis. Other essential ingredients are timefrequency analysis, models, and missing-data techniques. The strength of cancellation is that it can provide, in ideal conditions, infinite jammer rejection. Its weakness is that these ideal conditions may occur only locally, in time and/ or frequency, hence the need for models and missing-data techniques. This approach should benefit from future progress in signal processing techniques such as beamforming and ICA, and also from being cast into a systematic probabilistic framework. There are arguments to say that neural processing in natural organisms is in part based on cancellation. More basic knowledge is needed about the nature of these mechanisms, their anatomy and physiology, and the behavior that they allow.