Text normalization with varied data sources for conversational speech language modeling

by: S Schwarm, M Ostendorf

Acoustics, Speech, and Signal Processing, 2002. Proceedings. (ICASSP '02). IEEE International Conference on, Vol. 1 (2002), pp. I-789-I-792 vol.1.

View FullText article

IEEE Explore

Reviews [Write a review of this article]

The text from related topic but in different style was used in LM training for an ASR task. Text normalization was performed. Therefore, more training data could be used in LM training, and the OOV rate, as well as the recognition performance, was improved.

Reviewed by zzb3886 as

- 2008-06-25 01:41:47

Find related articles from these CiteULike users

zzb3886

Find related articles with these CiteULike tags

lm, normalization, text

Abstract

Collecting sufficient language model training data for good speech recognition performance in a new domain is often difficult. However, there may be other sources of data that are matched in terms of topic or style, if not both. This paper looks at the use of text normalization tools to make these data more suitable for language model training, in conjunction with mixture models to combine data from different sources. We specifically address the task of recognizing meeting speech, showing a small reduction in word error rate over a baseline language model trained from conversational speech data

@inproceedings{citeulike:2924657, abstract = {Collecting sufficient language model training data for good speech recognition performance in a new domain is often difficult. However, there may be other sources of data that are matched in terms of topic or style, if not both. This paper looks at the use of text normalization tools to make these data more suitable for language model training, in conjunction with mixture models to combine data from different sources. We specifically address the task of recognizing meeting speech, showing a small reduction in word error rate over a baseline language model trained from conversational speech data}, author = {Schwarm, S. and Ostendorf, M. }, booktitle = {Acoustics, Speech, and Signal Processing, 2002. Proceedings. (ICASSP '02). IEEE International Conference on}, citeulike-article-id = {2924657}, journal = {Acoustics, Speech, and Signal Processing, 2002. Proceedings. (ICASSP '02). IEEE International Conference on}, keywords = {lm, normalization, text}, pages = {I-789--I-792 vol.1}, posted-at = {2008-06-25 01:01:17}, priority = {0}, title = {Text normalization with varied data sources for conversational speech language modeling}, url = {http://ieeexplore.ieee.org/xpls/abs\_all.jsp?arnumber=1005858}, volume = {1}, year = {2002} }

RIS record

TY - CONF ID - citeulike:2924657 L3 - citeulike-article-id:2924657 TI - Text normalization with varied data sources for conversational speech language modeling T2 - Acoustics, Speech, and Signal Processing, 2002. Proceedings. (ICASSP '02). IEEE International Conference on JF - Acoustics, Speech, and Signal Processing, 2002. Proceedings. (ICASSP '02). IEEE International Conference on VL - 1 SP - I-789 EP - I-792 vol.1 N2 - Collecting sufficient language model training data for good speech recognition performance in a new domain is often difficult. However, there may be other sources of data that are matched in terms of topic or style, if not both. This paper looks at the use of text normalization tools to make these data more suitable for language model training, in conjunction with mixture models to combine data from different sources. We specifically address the task of recognizing meeting speech, showing a small reduction in word error rate over a baseline language model trained from conversational speech data KW - lm KW - normalization KW - text AU - Schwarm, S AU - Ostendorf, M PY - 2002/// UR - http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=1005858 ER -

RIS BibTeX