A single student model trained on essence knowledge from a teacher ensemble plus hard labels outperforms both a standard single model and the teacher ensemble itself on the Switchboard dataset.
Sequence distillation for purely sequence trained acoustic models,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2019 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Essence Knowledge Distillation for Speech Recognition
A single student model trained on essence knowledge from a teacher ensemble plus hard labels outperforms both a standard single model and the teacher ensemble itself on the Switchboard dataset.