Various Information Sources for HMM with Weighted Multiple Codebooks

Hacker C, Stemmer G, Steidl S, Nöth E, Niemann H (2003)

Publication Language: English

Publication Type: Conference contribution, Conference Contribution

Publication year: 2003

Original Authors: Hacker Christian, Stemmer Georg, Steidl Stefan, Nöth Elmar, Niemann Heinrich

Publisher: -

City/Town: -

Pages Range: 9-16

Conference Proceedings Title: Proceedings of the Speech Processing Workshop, Magdeburg, Germany, September 09

Event location: Magdeburg DE

URI: http://www5.informatik.uni-erlangen.de/Forschung/Publikationen/2003/Hacker03-VIS.pdf

Abstract

When semicontinuous HMM are used for acoustic modeling in speech recognition usually all states share a single code book. In our investigations we split the feature set into independent subsets and use separate codebooks for each part. This provides a higher modeling flexibility while keeping the parameter space compact. Further experiments integrate new information sources by using additional codebooks which are estimated in a supervised training. For instance codebooks for phone transitions are applied. Codebook exponents weight the different information sources. Relative reductions in word error rate up to 20 % have been achieved.

Authors with CRIS profile

Stefan Steidl Lehrstuhl für Informatik 14 (Bild- und Sprachverarbeitung) Elmar Nöth Professur für Informatik (Mustererkennung) Heinrich Niemann Technische Fakultät

How to cite

APA:

Hacker, C., Stemmer, G., Steidl, S., Nöth, E., & Niemann, H. (2003). Various Information Sources for HMM with Weighted Multiple Codebooks. In Wendemuth A. (Eds.), Proceedings of the Speech Processing Workshop, Magdeburg, Germany, September 09 (pp. 9-16). Magdeburg, DE: -: -.

MLA:

Hacker, Christian, et al. "Various Information Sources for HMM with Weighted Multiple Codebooks." Proceedings of the Speech Processing Workshop, Magdeburg Ed. Wendemuth A., -: -, 2003. 9-16.

BibTeX: Download