Model reference · open weights

japanese-avhubert-large_noise_pt

japanese-avhubert-large_noise_pt is an open-weight embedding model from enactic. japanese-avhubert-large_noise_pt (FP32) weighs 649 MB; the smallest configuration that runs it is RTX 3060 12 GB.

• japanese-avhubert-large_noise_pt is a 325M parameter feature-extraction model developed by enactic for audio-visual speech recognition.
• It is a self-supervised AVHuBERT variant pretrained on approximately 2,250 hours of Japanese data with noise augmentation to handle noisy environments.
• The model supports Japanese and is released under the cc-by-nc-4.0 license.

Summary of the enactic/japanese-avhubert-large_noise_pt model card, 2026-10-01

Hugging Face ↗Paper ↗