EERA-ASR: An Energy-Efficient Reconfigurable Architecture for Automatic Speech Recognition With Hybrid DNN and Approximate Computing.
Bo LiuHai QinYu GongWei GeMengwen XiaLongxing ShiPublished in: IEEE Access (2018)
Keyphrases
- automatic speech recognition
- reconfigurable architecture
- speech recognition
- word error rate
- speech signal
- conversational speech
- broadcast news
- hidden markov models
- speech retrieval
- systolic array
- wireless sensor networks
- spoken words
- word recognition
- speech corpus
- recognition errors
- acoustic features
- spontaneous speech
- speech recognizers
- noisy environments
- image processing
- machine learning