|
--- |
|
metrics: |
|
- cer |
|
--- |
|
|
|
## Introduction |
|
|
|
This repository provides the baseline model files for CNVSRC2023 (Chinese Continuous Visual Speech Recognition Challenge 2023). |
|
|
|
## Usage |
|
|
|
Please download these model files and use them in the [baseline code](https://github.com/MKT-Dataoceanai/CNVSRC2023Baseline). |
|
|
|
## Performance |
|
|
|
The following table shows these models' performance on their own tasks. |
|
| Training Data | Task | CER | File Name | |
|
|:-------------------------:|:----------------------:|:------:|:-----------------------------------------| |
|
| CN-CVS (<4s) | Pre-training | / | model_avg_14_23_cncvs_4s.pth | |
|
| CN-CVS (full) | Pre-training | / | model_avg_last10_cncvs_4s_30s.pth | |
|
| CN-CVS + CNVSRC-Single.Dev| Single-speaker VSR (T1)| 48.60% | model_avg_last5_cncvs_cnvsrc-single.pth | |
|
| CN-CVS + CNVSRC-Multi.Dev | Multi-speaker VSR (T2)| 58.37% | model_avg_last5_cncvs_cnvsrc-multi.pth | |
