Sixth European Conference on Speech Communication and Technology

Budapest, Hungary
September 5-9, 1999

Semi-Supervised Adaptation of Acoustic Models for Large-Volume Dictation

Colin W. Wightman (1), Ted A. Harder (2)

(1) Department of Computer Science, Minnesota State University, Mankato, MN, USA
(2) Department of Electrical Engineering, Duke University, Durham, NC, USA

Using a Large-Vocabulary, Continuous Speech Recognizer in a high-Volume application such as a commercial transcription service presents a different set of challenges and constraints than in a laboratory setting. We examine these differences with regard to acoustic model adaptation and find serious shortcomings in both the supervised and unsupervised approaches. We then examine a new method, semi-supervised adaptation, which overcomes the limitations of the other methods and can reduce recognition error rates by as much as 15% more than the reduction obtained through unsupervised adaptation.

Full Paper (PDF)   Gnu-Zipped Postscript

Bibliographic reference.  Wightman, Colin W. / Harder, Ted A. (1999): "Semi-supervised adaptation of acoustic models for large-volume dictation", In EUROSPEECH'99, 1371-1374.