Hypothesis Stitcher for End-to-End Speaker-Attributed ASR on Long-Form Multi-Talker Recordings.
Xuankai ChangNaoyuki KandaYashesh GaurXiaofei WangZhong MengTakuya YoshiokaPublished in: ICASSP (2021)
Keyphrases
- end to end
- automatic speech recognition
- speech recognition
- wireless ad hoc networks
- admission control
- ad hoc networks
- multipath
- congestion control
- high bandwidth
- acoustic features
- speaker verification
- content delivery
- multi hop
- audio visual
- spontaneous speech
- hidden markov models
- real time
- text localization and recognition
- internet protocol
- language model
- image sequences