Sixth European Conference on Speech Communication and Technology
(EUROSPEECH'99)

Budapest, Hungary
September 5-9, 1999

Concurrent Speakers Separation Through Binaural Processing of Stereo Recordings

Joaquin Gonzalez-Rodriguez, Santiago Cruz-Llanas, Javier Ortega-Garcia

DIAC-EUIT.Telecomunicacion - Universidad Politecnica de Madrid, Campus Sur, Madrid, Spain

In this contribution, we address the problem of speech separation from concurrent speakers through the use of a binaural processor. In this system, we take advantage of both ITD’s (Inter-aural Time Differences) and ILD’s (Inter-aural Level Difs.) present in the binaural input signals to perform spatial localisation and discrimination. However, in applications as Forensic Acoustics where hidden recordings are desired we cannot make use of head simulators, but we can also obtain ITD’s and ILD’s through stereo recordings with a cardioid microphone pair. In this way, after a new training of the processor, we are able to extract the information arriving from the desired direction from a simple stereo recording. Several experiments with static and moving speakers have been performed, and some audio results on concurrent speakers separation are also included.


Full Paper (PDF)   Gnu-Zipped Postscript

Bibliographic reference.  Gonzalez-Rodriguez, Joaquin / Cruz-Llanas, Santiago / Ortega-Garcia, Javier (1999): "Concurrent speakers separation through binaural processing of stereo recordings", In EUROSPEECH'99, 2595-2598.