Output details
11 - Computer Science and Informatics
University of Sheffield
Article title
Error Approximation and Minimum Phone Error Acoustic Model Estimation
Type
D - Journal article
Title of journal
IEEE Transactions on Audio, Speech, and Language Processing
Article number
-
Volume number
18
Issue number
6
First page of article
1269
ISSN of journal
15587916
Year of publication
2010
URL
-
Number of additional authors
1
Additional information
<22> State-of-the-art speech recognition systems employ a form of training that requires counting recognition errors. However, exact error counts are prohibitively expensive to compute. This paper presents a novel error approximation technique that performs better than any method previously used. Improved error approximation was shown to translate into consistently better speech recognition performance. The method has been employed in the Kaldi speech recognition toolkit and led to a project funded by CISCO Systems (Unsupervised Domain Adaptation) to develop speech transcription for teleconferencing systems.
Interdisciplinary
-
Cross-referral requested
-
Research group
None
Citation count
7
Proposed double-weighted
No
Double-weighted statement
-
Reserve for a double-weighted output
No
Non-English
No
English abstract
-
