Compression transparent low-level description of audio signals

Abstract

A new low level audio descriptor that represents the psychoacoustic noise floor shape of an audio frame is proposed. Results presented indicate that the proposed descriptor is far more resilient to compression noise than any of the MPEG-7 low level audio descriptors. In fact, across a wide range of files, on average the proposed scheme fails to uniquely identify only five frames in every ten thousand. In addition, the proposed descriptor maintains a high resilience to compression noise even when decimated to use only one quarter of the values per frame to represent the noise floor. This characteristic indicates the proposed descriptor presents a truly scalable mechanism for transparently describing the characteristics of an audio frame.