README.md

---
language:
- en
tags:
- audio-visual
- speech-recognition
- mcorec
- chime-9
- avsr
library_name: pytorch
pipeline_tag: automatic-speech-recognition
---

# MCoRec D2-STREAM (Nemotron chunked visual)

Checkpoint from the MCoRec / CHiME-9 Task 1 AVSR project closeout.

## Evaluation

- **Split:** official confirmation set (20 sessions / 110 speakers)
- **Protocol:** chunked visual streaming conditioned ASR
- **Confirm WER:** `0.4865`

## Files

Weights are uploaded as released training artifacts (`.ckpt` / `.nemo` / `.pt`).
Use the corresponding project inference stack to load them.

## Collection

Part of [`jtygarfield/avsr-project`](https://huggingface.co/collections/jtygarfield/avsr-project).