Efficient Minimum Word Error Rate Training of RNN-Transducer for End-to-End Speech Recognition.
Jinxi GuoGautam TiwariJasha DroppoMaarten Van SegbroeckChe-Wei HuangAndreas StolckeRoland MaasPublished in: INTERSPEECH (2020)
Keyphrases
- speech recognition
- end to end
- word error rate
- automatic speech recognition
- language model
- hidden markov models
- isolated word
- handwriting recognition
- speech signal
- speech synthesis
- error rate
- recurrent neural networks
- noisy environments
- multi modal
- speaker identification
- speech recognizer
- pattern recognition
- acoustic models
- speech recognition systems
- image processing
- neural network
- maximum likelihood
- broadcast news
- feature space