NASA Logo

NTRS

NTRS - NASA Technical Reports Server

Back to Results
Improving Acoustic Models by Watching TelevisionObtaining sufficient labelled training data is a persistent difficulty for speech recognition research. Although well transcribed data is expensive to produce, there is a constant stream of challenging speech data and poor transcription broadcast as closed-captioned television. We describe a reliable unsupervised method for identifying accurately transcribed sections of these broadcasts, and show how these segments can be used to train a recognition system. Starting from acoustic models trained on the Wall Street Journal database, a single iteration of our training method reduced the word error rate on an independent broadcast television news test set from 62.2% to 59.5%.
Document ID
19990045639
Document Type
Reprint (Version printed in journal)
Authors
Witbrock, Michael J. (Carnegie-Mellon Univ. Pittsburgh, PA United States)
Hauptmann, Alexander G. (Carnegie-Mellon Univ. Pittsburgh, PA United States)
Date Acquired
August 19, 2013
Publication Date
March 19, 1998
Subject Category
Communications and Radar
Report/Patent Number
AD-A350494
CMU-CS-98-110
Funding Number(s)
CONTRACT_GRANT: NSF IRI-94-11299
Distribution Limits
Public
Copyright
Work of the US Gov. Public Use Permitted.
Document Inquiry