From IntRoLab
Revision as of 21:08, 10 November 2008 by Dominic (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)

Return to AUDIBLE

Microphone Array Source Separation

The following demonstrates the use of a microphone array for separating three simultaneous speech sources in presence of both reverberation and background noise.

Signal at microphone output

The signals below is taken directly from one of the microphones in the array. It shows all three sources, mixed with background noise.

AUDIBLE MicrophoneOutput.jpg

Signal at microphone output (wav)

Separated sources after Geometric Source Separation (GSS) only

The signals shown here demonstrates the capabilities of the Geometric Source Separation (GSS) algorithm we use. While interference can still be heard, its level is lower than for a single microphone.

AUDIBLE Separated1.jpg AUDIBLE Separated2.jpg AUDIBLE Separated3.jpg
Source 1 (wav) Source 2 (wav) Source 3 (wav)

Separated sources after Geometric Source Separation (GSS) and post-filter

The signals presented here is the result obtained after applying our post-filter to the output of the GSS algorithm. Our post-filter is able to remove most of the remaining background noise and interferences at the expense of minor distortion to the source of interest.

AUDIBLE SeparatedFiltered1.jpg AUDIBLE SeparatedFiltered2.jpg AUDIBLE SeparatedFiltered3.jpg
Source 1 (wav) Source 2 (wav) Source 3 (wav)

Reference (original) sources

These signals are provided here for reference only and correspond to the real (unknown to the algorithm) sources.

AUDIBLE Reference1.jpg AUDIBLE Reference2.jpg AUDIBLE Reference3.jpg
Source 1 (wav) Source 2 (wav) Source 3 (wav)