Speech identification using a sequence-based heuristic

This 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.