BERT Meets CTC: New Formulation of End-to-End Speech Recognition with Pre-trained Masked Language Model.
Yosuke HiguchiBrian YanSiddhant AroraTetsuji OgawaTetsunori KobayashiShinji WatanabePublished in: EMNLP (Findings) (2022)
Keyphrases
- end to end
- speech recognition
- language model
- pre trained
- language modeling
- training data
- probabilistic model
- information retrieval
- n gram
- query expansion
- test collection
- document retrieval
- automatic speech recognition
- retrieval model
- mixture model
- word error rate
- speech signal
- training examples
- query terms
- hidden markov models
- active learning
- image sequences
- neural network
- multi modal
- machine translation