Home

Blind extraction of dominant target sources

This page demonstrates sound examples for blind extraction of dominant target sources.

Experimental setups

Sound examples

Single-target cases

1 dominant target speech + 6 background interference speeches

Target position a120b120c120c170
Mixture x1x1 x1x1
Separation only with ICA y1y1 y1y1
Separation by ICA and T-F masking (fewer zeros) y1y1 y1y1
Separation by ICA and T-F masking (moderate) y1y1 y1y1
Separation by ICA and T-F masking (more zeros) y1y1 y1y1

Multi-target case

3 dominant target speeches + 6 background interference speeches

Target position a120b120c170
Mixture x1
Separation only with ICA y1y2 y3
Separation by ICA and T-F masking (moderate) y1 y2 y3

Reference

  1. H. Sawada, S. Araki, R. Mukai, S. Makino, "Blind Extraction of Dominant Target Sources Using ICA and Time-Frequency Masking," IEEE Trans. Audio, Speech, and Language Processing, vol.14, no.6, pp.2165-2173, Nov. 2006 (Paper: PDF)