Model reference · open weights

hear-pytorch

hear-pytorch is an open-weight embedding model from Google. hear-pytorch (BF16) weighs 1.2 GB; the smallest configuration that runs it is RTX 3060 12 GB.

• HeAR is a health acoustic foundation model developed by Google that generates 512-dimensional embeddings from two-second audio clips to represent non-semantic respiratory sounds.
• The model uses a ViT-L architecture trained on a large unlabelled corpus to support downstream tasks such as disease screening and lung function monitoring.
• It is available as a PyTorch implementation under the Health AI Developer Foundations terms of use.

Summary of the google/hear-pytorch model card, 2026-10-01

Hugging Face ↗GitHub ↗Paper ↗