Heinrich, G.G.Heinrich2022-03-102022-03-102005https://publica.fraunhofer.de/handle/publica/34904010.1109/ELMAR.2005.1936832-s2.0-33745265680This article shows an approach to identify speech in a given signal with moderate noise levels. It is based on the segmental structure of speech processes and has two advantages: It is trained exclusively from speech sounds and has a relatively low computational footprint at testing time.ensignal processingSpeech Inputspeech detectionsignal classificationperceptual linear predictionsupport vector machine006conference paper