Computational auditory scene analysis-constrained array processing for sound source separation

Abstract

In this work, techniques are developed and studied for the extraction of single-source acoustic signals out of multi-source signals. Such extracted signals can be used in a variety of applications including: automatic speech recognition, teleconferencing, and robot auditory systems. Most previous approaches fall into two categories: computational auditory scene analysis (CASA) and array signal processing. The approach taken here is to combine these complementary techniques into an integrated one: CASA-constrained array processing. In principle, this integrated approach should provide a performance gain since the information used by array processing (direction of propagation through a sound-field) is independent of other CASA features (fundamental frequency, on/offset, etc.). One difficulty encountered by CASA that can be overcome by array processing is the sequential grouping of spectrally dissimilar phonemes in a speech signal, such as a fricative followed by a vowel. The method presented here differs from standard array processing by the addition of CASA features for the signal separation decision. Compared to other CASA systems that use binaural cues, it: (1) is not limited to two microphones (since the goal is not auditory system modeling); and (2) makes complete use of source location and other CASA features—for simultaneous and sequential grouping.