A distributed architecture for fast SGD sequence discriminative training of DNN acoustic models.
George SaonPublished in: SLT (2014)
Keyphrases
- discriminative training
- distributed architecture
- acoustic models
- hidden markov models
- maximum likelihood
- probabilistic interpretation
- posterior probability
- automatic speech recognition
- speech recognition
- support vector machine
- multi robot
- training process
- data sets
- broadcast news
- natural language processing
- bayesian framework
- training data
- machine learning
- speech recognizer
- neural network