Computational auditory scene analysis: A representational approach

This thesis addresses the problem of how a listener groups together acoustic components which have arisen from the same environmental event, a phenomenon known auditory scene analysis. A computational model of auditory scene analysis is, which is able to separate speech from a variety of interfering noises. model consists of four processing stages. Firstly, the auditory periphery is by a bank of bandpass filters and a model of inner hair cell function. In second stage, physiologically-inspired models of higher auditory organization - maps - are used to provide a rich representational basis for scene analysis. in the acoustic input are coded by an ant ocorrelation map and a crosscorrelation map. Information about spectral continuity is extracted by a frequency map. The times at which acoustic components start and stop are identified an onset map and an offset map. the third 8tage of processing, information from the periodicity and frequency maps is used to characterize the auditory scene as a collection of symbolic auditory objects. Finally, a search strategy identifies objects that have similar and groups them together. Specifically, objects are likely to form a group they have a similar periodicity, onset time or offset time. model has been evaluated in two ways, using the task of segregating voiced from a number of interfering sounds such as random noise, "cocktail party" and other speech. Firstly, a waveform can be resynthesized for each group the auditory scene, so that segregation performance can be assessed by informal tests. The resynthesized speech is highly intelligible and fairly natural. , the linear nature of the resynthesis process allows the signal-to-noise ratio (SNR) to be compared before and after segregation. An improvement in SNR is after segregation for each type of interfering noise. Additionally, the performance of the model is significantly better than that of a conventional frame-based segregation strategy.