Multi-Frame Cross-Entropy Training for Convolutional Neural Networks in Speech Recognition.
Tom SercuNeil MallinarPublished in: CoRR (2019)
Keyphrases
- speech recognition
- cross entropy
- multi frame
- convolutional neural networks
- language model
- language modeling
- acoustic models
- motion estimation
- hidden markov models
- log likelihood
- point correspondences
- pattern recognition
- image sequences
- super resolution
- motion analysis
- speech recognizer
- training set
- maximum likelihood
- speech recognition systems
- optical flow
- optic flow
- mixture model
- training process
- filtering algorithm
- supervised learning
- computationally efficient
- evaluation metrics
- image processing
- information retrieval
- machine learning