DistilHuBERT
A lightweight distilled variant of HuBERT that applies knowledge distillation to compress the self-supervised speech representation learning model into a smaller-parameter, faster-inference version. It is commonly used for feature extraction in downstream tasks such as speech classification.
Timeline (last 90 days)
DistilHuBERT is a lightweight distilled version of HuBERT (Hidden-Unit BERT), compressed via knowledge distillation into a version with fewer parameters and faster inference.
The study uses L2-ARCTIC and multiple external English accent datasets to compare two evaluation settings: file-level splitting and speaker-isolated splitting.
All Facts (2)
DistilHuBERT is a lightweight distilled version of HuBERT (Hidden-Unit BERT), compressed via knowledge distillation into a version with fewer parameters and faster inference.
50%UnverifiedThe study uses L2-ARCTIC and multiple external English accent datasets to compare two evaluation settings: file-level splitting and speaker-isolated splitting.
50%