Robust Voice Activity Detection for Distributed Microphones by Modeling of Power Ratios
Konferenz: Sprachkommunikation 2010 - 9. ITG-Fachtagung
06.10.2010 - 08.10.2010 in Bochum, Deutschland
Tagungsband: Sprachkommunikation 2010
Seiten: 4Sprache: EnglischTyp: PDFPersönliche VDE-Mitglieder erhalten auf diesen Artikel 10% Rabatt
Matheja, Timo; Buck, Markus (Nuance Communications Aachen GmbH, Ulm, Söflinger Str. 100, 89077 Ulm, Germany)
In this paper a method for a robust frequency selective voice activity detection (VAD) is presented that evaluates the power ratios between several distributed microphones. The method refers to a setup where several speakers each have a dedicated microphone. In many acoustic environments, particularly in a car cabin, strong reflections may occur depending on the room acoustics. Due to destructive interference at certain frequencies the dedicated microphone may show a weaker speech signal component than a distant microphone. This effect impairs standard VAD methods. With the proposed approach a threshold is adapted by online estimation of the distribution of the particular power ratios. Therewith a speech detection can be obtained even in those subbands where the described power inversions occur.