Speech enhancement based on the response features of facilitated EI neurons

Authors:
André B. Cavalcante;Danilo P. Mandic;Tomasz M. Rutkowski;Allan Kardec Barros
Affiliations:
Laboratory for Biological Information Processing, Universidade Federal do Maranhāo, Brazil;Department of Electrical and Electronic Engineering, Imperial College London, United Kingdom;Laboratory for Advanced Brain Signal Processing, Brain Science Institute Riken, Japan;Laboratory for Biological Information Processing, Universidade Federal do Maranhāo, Brazil
Venue:
ICA'06 Proceedings of the 6th international conference on Independent Component Analysis and Blind Signal Separation
Year:
2006

Citing 3
Cited 0

A fast fixed-point algorithm for independent component analysis

Neural Computation
Single channel speech enhancement by efficient coding

Signal Processing
Estimation of speech embedded in a reverberant and noisy environment by independent component analysis and wavelets

IEEE Transactions on Neural Networks

Quantified Score

Hi-index	0.00

Visualization

Abstract

A real-time approach for the enhancement of speech at zero degree azimuth is proposed. This is achieved inspired by the response features of the “Facilitated EI neurons”. This way, frequency segregation through a bandpass filter bank is followed by “supression analysis” which inhibits sources that are not at “facilitated” positions. Unlike with the existing approaches for the solution of cocktail party problem, where the performance under low SNR (signal-to-noise ratio) reverberation conditions is severely limited, the proposed approach has the capability to circumvent these problems. This is quantified through both objective and subjective performance measures and supported by real world simulation examples.