Efficient reduction of gaussian components using MDL criterion for HMM-based speech recognition

ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings

Volumn 1, Issue , 2002, Pages 869-872

  Shinoda, Koichi ^a   Iso, Ken Ichi ^a  

^a NEC CORPORATION   (Japan)

Author keywords

[No Author keywords available]

Indexed keywords

GAUSSIAN COMPONENTS; HIDDEN MARKOV MODEL; MINIMUM DESCRIPTION LENGTH;

MARKOV PROCESSES; MATHEMATICAL MODELS; OPTIMIZATION; PROBABILITY DENSITY FUNCTION; TREES (MATHEMATICS);

SPEECH RECOGNITION;

EID: 0036305005     PISSN: 15206149     EISSN: None     Source Type: Journal    
DOI: 10.1109/ICASSP.2002.5743877     Document Type: Article

References (11)

1. Acoustic modeling based on the MDL criterion for speech recognition
2. MDL-based context-dependent subword modeling for speech recognition
3. A study on robustness against data insufficiency for speech recognition
4. Model selection in acoustic modeling
5. Dynamic threshold setting via bayesian information criterion (BIC) in HMM training
6. Speech recognition using tree-structured probability density function
7. Universal coding, information, prediction, and estimation
8. Dynamically configurable acoustic models for speech recognition
9. Pruning of state-tying tree using Bayesian information criterion with multiple mixtures
10. Parameter typing and Gaussian clustering for faster, better, and smaller speech recognition
11. Mixture trees - Hierarchically tied mixture densities for modeling HMM emission probabilities