When a mismatch can be good

Subscribe (Full Service)


	Search: The ACM Digital Library The Guide

	Feedback

When a mismatch can be good: large vocabulary speech recognition trained with idealized tandem features

Full text

Pdf (105 KB)

Source

Symposium on Applied Computing archive
Proceedings of the 2008 ACM symposium on Applied computing table of contents

Fortaleza, Ceara, Brazil

SESSION: Natural language processing and speech recognition table of contents

Pages 1574-1577

Year of Publication: 2008

ISBN:978-1-59593-753-7

Authors

Arlo Faria	University of California at Berkeley, Berkeley, CA
Nelson Morgan	International Computer Science Institute, Berkeley, CA

Sponsor

SIGAPP: ACM Special Interest Group on Applied Computing

Publisher

ACM New York, NY, USA

Bibliometrics

Downloads (6 Weeks): 7, Downloads (12 Months): 21, Citation Count: 1

Additional Information:

abstract references cited by index terms collaborative colleagues

Tools and Actions:	Request Permissions Review this Article Save this Article to a Binder Display Formats: BibTeX EndNote ACM Ref

DOI Bookmark:	Use this link to bookmark this Article: http://doi.acm.org/10.1145/1363686.1364055 What is a DOI?

ABSTRACT

This paper explores Tandem feature extraction used in a large-vocabulary speech recognition system. In this framework a multi-layer perceptron estimates phone probabilities which are treated as acoustic observations in a traditional HMM-GMM system. To determine a lower error bound, we simulated an idealized classifier based on alignment of reference transcriptions. This cheating experiment demonstrated a best-case scenario for Tandem feature extraction, highlighting the potential for dramatic system improvement. More importantly, we discovered a way to exploit the result without cheating: using the simulated classifier during training and a MLP classifier at test, the performance improved despite the mismatched Tandem features.

REFERENCES

Note: OCR errors may be found in this Reference List extracted from the full text article. ACM has opted to expose the complete List rather than only correct and linked references.

	1	H. Hermansky, D. Ellis, and S. Sharma. Tandem connectionist feature extraction for conventional HMM systems. Proc. ICASSP, 2000.
	2	M. Hwang, X. Lei, W. Wang, and T. Shinozaki. Investigation on Mandarin Broadcast News Speech Recognition. Proc. Interspeech, 2006.
	3	X. Lei, M. Siu, M.-Y. Hwang, M. Ostendorf, and T. Lee. Improved Tone Modeling for Mandarin Broadcast News Speech Recognition. Proc. Interspeech, 2006.
	4	G. Peng, M.-Y. Hwang, and M. Ostendorf. Automatic acoustic segmentation for speech recognition on broadcast recordings. Proc. Interspeech, 2007.
	5	D. Povey, B. Kingsbury, L. Mangu, G. Saon, H. Soltau, and G. Zweig. fMPE: Discriminatively Trained Features for Speech Recognition. Proc. ICASSP, 2005.
	6	D. Povey and P. Woodland. Minimum phone error and I-smoothing for improved discriminative training. Proc. ICASSP, 2002.
	7	J. Zheng, O. Cetin, M.-Y. Hwang, X. Lei, A. Stolcke, and N. Morgan. Combining Discriminative Feature, Transform, and Model Training for Large Vocabulary Speech Recognition. Proc. ICASSP, 2007.
	8	Q. Zhu, A. Stolcke, B. Chen, and N. Morgan. Using MLP features in SRI's conversational speech recognition system. Proc. Interspeech, 2005.