An Evaluation of Discriminative Training for Hidden Markov Models in a Real-Environment Speech-Oriented Guidance System
スポンサーリンク
概要
- 論文の詳細を見る
This paper presents experimental evaluations of discriminative training of the acoustic model in a real-environment speech-oriented guidance system. Recently, discriminative training techniques have made significant progress in automatic speech recognition (ASR) based on the hidden Markov model (HMM). Their effectiveness has been confirmed in well-known ASR data sets. It is also worthwhile to investigate its effectiveness in more challenging speech data sets. In this paper, we evaluate the effectiveness of discriminative training in speech data recorded in a real environment speech-oriented guidance system, "Takemaru-kun", which has been installed in a public place in November 2002 and has recorded input speech data since then. The recorded speech data include very spontaneous speech of various speakers such as children, adults, and elderly people. Maximum Mutual Information (MMI) training is implemented for building HMMs using these speech data. First we investigate the performance of discriminative training by changing initial conditions in the acoustic model structure and lattice generation. Then, we optimize several training parameters such as the acoustic scale factor and the I-smoothing parameter. Our results show that MMI training yields around 2% absolute word accuracy improvement compared with ML training in the speech data recorded in the "Takemaru-kun" system.
- 2010-07-15
著者
-
Denis Babani
Nara Institute of Science and Technology
-
Tomoki Toda
Nara Institute of Science and Technology
-
Hiroshi Saruwatari
Nara Institute of Science and Technology
-
Kiyohiro Shikano
Nara Institute of Science and Technology
関連論文
- Stacked Generalization for Topic Classification of Spoken Inquiries
- The Use of Transformed Normal Speech Data in Acoustic Model Training for Non-Audible Murmur Recognition
- An Evaluation of Discriminative Training for Hidden Markov Models in a Real-Environment Speech-Oriented Guidance System
- Inquiry Classification in a Speech-Oriented Guidance System Using Discriminative Learning
- Comparison of Methods for Topic Classification of Spoken Inquiries (Preprint)