On the equivalence of Gaussian and log-linear HMMs.
Georg HeigoldPatrick LehnenRalf SchlüterHermann NeyPublished in: INTERSPEECH (2008)
Keyphrases
- log linear
- discriminative training
- hidden markov models
- maximum likelihood
- log linear models
- probabilistic modeling
- efficient learning
- latent variables
- speech recognition
- probabilistic interpretation
- multi class
- feature selection
- markov networks
- conditional random fields
- expectation maximization
- linguistically motivated
- bayesian networks