SLAM: A Unified Encoder for Speech and Language Modeling via Speech-Text Joint Pre-Training.
Ankur BapnaYu-An ChungNan WuAnmol GulatiYe JiaJonathan H. ClarkMelvin JohnsonJason RiesaAlexis ConneauYu ZhangPublished in: CoRR (2021)
Keyphrases
- language modeling
- language model
- speech recognition
- information retrieval
- finite state transducers
- word error rate
- retrieval model
- probabilistic model
- speech signal
- query expansion
- cross lingual
- text retrieval
- automatic speech recognition
- training set
- n gram
- feature vectors
- knn
- feature space
- retrieval effectiveness
- machine learning