Ninth International Conference on Spoken Language Processing

Pittsburgh, PA, USA
September 17-21, 2006

Detection and Separation of Speech Events in Meeting Recordings

Futoshi Asano, Jun Ogata

AIST, Japan

When applying automatic speech recognition (ASR) to meeting recordings including spontaneous speech, the performance of ASR is greatly reduced by the overlap of speech events. In this paper, a method of separating the overlapping speech events using an adaptive beamforming (ABF) framework is proposed. The main feature of this method is that all the necessary information for the adaptation of ABF, including microphone calibration, is obtained from meeting recordings based on the results of speech event detection. The performance of the separation is evaluated via ASR using real meeting recordings.

Full Paper

Bibliographic reference.  Asano, Futoshi / Ogata, Jun (2006): "Detection and separation of speech events in meeting recordings", In INTERSPEECH-2006, paper 1098-Thu2FoP.3.