Model
—
Frame-VAD Multilingual MarbleNetFrame-based VAD model using MarbleNet model trained with real multilingual data and synthetic English data.
Use the NGC CLI to download:
Copied!
1 Version
1.20.0Selected05/02/2023 6:40 PM UTC490 KB Copied!
1.20.0Selected
05/02/2023 6:40 PM UTC490 KB
Copied!
Accuracy
| Key | Value |
|---|---|
| AMI AUROC | 95.83 |
| CH109 AUROC | 92.43 |
| AVA AUROC | 93.77 |
| VoxConv AUROC | 96.45 |
Model
| Key | Value |
|---|---|
| Architecture | MarbleNet-3x2x64-20ms |
| Outputs | Sequence of speech probability for each 20ms frame |
| Inputs | Audio |
| Number of Weights | 94K |