A compressed HuBERT can be trained with the original masked-prediction objective using k-means labels from the teacher, beating feature-distillation methods on four SUPERB tasks.
teacher” model to a smaller “student
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
DiceHuBERT: Distilling HuBERT with a Self-Supervised Learning Objective
A compressed HuBERT can be trained with the original masked-prediction objective using k-means labels from the teacher, beating feature-distillation methods on four SUPERB tasks.