mcorec-d2-stream / README.md
jtygarfield's picture
Add d2_stream_chunked sealed confirm checkpoint
45e9596 verified
|
Raw
History Blame Contribute Delete
706 Bytes
---
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).