TODE: A Decoder for Continuous Speech Recognition
Darren C. Moore · Infoscience (Ecole Polytechnique Fédérale de Lausanne) · 2002
This document describes a new continuous speech decoder, TODE, which is compatible with the Torch machine learning software library. A brief theory of speech recognition is presented followed by a detailed description of the architecture of TODE and the components used in its implementation. Results are presented from experiments that assessed the performance of TODE and compared it with two other commonly-used decoders.