---
license: apache-2.0
---
# EgoSpeak Checkpoints

This repo contains final checkpoints for the EgoSpeak project.  
- **Models**: `lstr`, `mamba`, `miniroad`  
- **Datasets**: `easycom`, `ego4dshuffle`, `ytconv_pretrained`  
- **Modalities**: `A` (audio), `V` (video), `AV` (audio+video)  

## File Naming Convention
- `mamba_easycom_A.pth` → Mamba model, trained on EasyCom, audio-only