Training dynamic models using early exits for automatic speech recognition on resource-constrained devices.
George August WrightUmberto CappellazzoSalah ZaiemDesh RajLucas Ondel YangDaniele FalavignaAlessio BruttiPublished in: CoRR (2023)
Keyphrases
- resource constrained
- dynamic model
- automatic speech recognition
- speech recognition
- resource constraints
- embedded systems
- sensor networks
- experimental data
- speech signal
- wireless sensor networks
- acoustic models
- hidden markov models
- broadcast news
- word error rate
- spoken words
- word recognition
- multipath
- rfid tags
- speech retrieval
- speech corpus
- recognition errors
- conversational speech
- multiple models
- spontaneous speech
- data sets
- low cost
- wireless networks
- temporal constraints
- speech synthesis
- speech sounds