First European Conference on Speech Communication and Technology

Paris, France
September 27-29, 1989

Modeling Between-Word Coarticulation in Continuous Speech Recognition

Mei-Yuh Hwang, Hsiao-Wuen Hon, Kai-Fu Lee

School of Computer Science, Carnegie Mellon University, Pittsburgh, PA, USA

This paper describes the addition of between-word coarticulation modeling into sphinx, an accurate large-vocabulary speaker-independent continuous speech recognition system. Between-word coarticulation is a major source of phonetic variability in continuous speech. By detailed modeling of between-word triphones and utilizing the generalized triphone technique, we obtain an error rate reduction of 16% to 29% for different test sets and grammars on the DARPA Resource Management (RM) Task.

Full Paper

Bibliographic reference.  Hwang, Mei-Yuh / Hon, Hsiao-Wuen / Lee, Kai-Fu (1989): "Modeling between-word coarticulation in continuous speech recognition", In EUROSPEECH-1989, 1005-1008.