Upload folder using huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,34 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
library_name: mlx
|
| 4 |
+
tags:
|
| 5 |
+
- speech-enhancement
|
| 6 |
+
- audio
|
| 7 |
+
- mlx
|
| 8 |
+
---
|
| 9 |
+
|
| 10 |
+
# MossFormer2 SE 6-bit (MLX)
|
| 11 |
+
|
| 12 |
+
48kHz speech enhancement model converted to MLX format (6-bit quantized).
|
| 13 |
+
|
| 14 |
+
## Original Model
|
| 15 |
+
[alibabasglab/MossFormer2_SE_48K](https://huggingface.co/alibabasglab/MossFormer2_SE_48K)
|
| 16 |
+
|
| 17 |
+
## Usage
|
| 18 |
+
|
| 19 |
+
```python
|
| 20 |
+
from mlx_audio.sts.models.mossformer2_se import MossFormer2SEModel
|
| 21 |
+
|
| 22 |
+
model = MossFormer2SEModel.from_pretrained("starkdmi/MossFormer2-SE-6bit")
|
| 23 |
+
enhanced = model.enhance("noisy.wav")
|
| 24 |
+
```
|
| 25 |
+
|
| 26 |
+
## Precision Variants
|
| 27 |
+
- [MossFormer2-SE](https://huggingface.co/starkdmi/MossFormer2-SE) (fp32, 211MB)
|
| 28 |
+
- [MossFormer2-SE-fp16](https://huggingface.co/starkdmi/MossFormer2-SE-fp16) (fp16, 106MB)
|
| 29 |
+
- [MossFormer2-SE-8bit](https://huggingface.co/starkdmi/MossFormer2-SE-8bit) (int8, 86MB)
|
| 30 |
+
- [MossFormer2-SE-6bit](https://huggingface.co/starkdmi/MossFormer2-SE-6bit) (int6, 75MB) ← this
|
| 31 |
+
- [MossFormer2-SE-4bit](https://huggingface.co/starkdmi/MossFormer2-SE-4bit) (int4, 64MB)
|
| 32 |
+
|
| 33 |
+
## Performance
|
| 34 |
+
~30x real-time on Apple M4
|