Deleted strategy for MMI-based HMM training

Author

Kim, Nam Soo ; Un, Chong Kwan

Author_Institution

Dept. of Electr. Eng., Seoul Nat. Univ., South Korea

Volume

6

Issue

3

fYear

1998

fDate

5/1/1998 12:00:00 AM

Firstpage

299

Lastpage

303

Abstract

We apply the maximum mutual information (MMI) criterion to discriminative training of hidden Markov model (HMM) parameters. In contrast to the conventional MMI training approach, we adopt the cross-validatory strategy with which the parameters are estimated on a part and assessed on the other parts of the training data. For this purpose, we propose the deleted MMI training method, which performs cross-validatory parameter updating while maintaining the converging behavior of the conventional MMI-based algorithm. The proposed method is compared to the conventional MMI approach in classification of artificial data and in speaker-independent continuous speech recognition, and shows better performance

Keywords

hidden Markov models; information theory; parameter estimation; speech recognition; HMM parameters; MMI-based HMM training; MMI-based algorithm; artificial data classification; converging behavior; cross-validatory parameter updating; deleted MMI training method; discriminative training; hidden Markov model; maximum mutual information; parameter estimation; performance; speaker-independent continuous speech recognition; training data; Computational complexity; Dynamic programming; Gaussian processes; Hidden Markov models; Kernel; Natural languages; Parameter estimation; Signal processing algorithms; Speech processing; Speech recognition;

fLanguage

English

Journal_Title

Speech and Audio Processing, IEEE Transactions on

Publisher

ieee

ISSN

1063-6676

Type

jour

DOI

10.1109/89.668824

Filename

668824