Robust automatic transcription of lectures

Abstract

Automatic transcription of lectures is becoming an important task. Possible applications can be found in the fields of automatic translation or summarization, information retrieval, digital libraries, education and communication research. Ideally those systems would operate on distant recordings, freeing the presenter from wearing body-mounted microphones. This task, however, is surpassingly difficult, given that the speech signal is severely degraded by background noise and reverberation.