All-phoneme ergodic hidden Markov network for unsupervised speaker adaptation.
Yasunaga MiyazawaJun-ichi TakamiShigeki SagayamaShoichi MatsunagaPublished in: ICASSP (1) (1994)
Keyphrases
- markov networks
- speaker adaptation
- maximum likelihood
- speech recognition
- speaker dependent
- graphical models
- em algorithm
- unsupervised learning
- supervised learning
- automatic speech recognition
- speaker independent
- expectation maximization
- hidden variables
- first order logic
- bayesian inference
- posterior probability
- bayesian networks
- hidden markov models
- maximum a posteriori
- mixture model
- semi supervised
- gaussian mixture model
- belief propagation
- random variables
- probabilistic model
- speaker identification
- conditional random fields
- speech signal
- speech recognizer
- low level
- pattern recognition