Hierarchical Bayesian combination of plug-in maximum a posteriori decoders in deep neural networks-based speech recognition and speaker adaptation.
Zhen HuangSabato Marco SiniscalchiChin-Hui LeePublished in: Pattern Recognit. Lett. (2017)
Keyphrases
- maximum a posteriori
- speaker adaptation
- speech recognition
- maximum likelihood
- neural network
- pattern recognition
- markov random field
- speaker independent
- image reconstruction
- speech recognizer
- em algorithm
- speaker dependent
- automatic speech recognition
- bayesian framework
- posterior distribution
- hyperparameters
- hidden markov models
- expectation maximization
- noisy environments
- speech synthesis
- language model
- energy function
- speech signal
- speech recognition systems
- mixture model
- belief propagation
- conditional random fields
- gaussian distribution