Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,26 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: cc-by-nc-sa-4.0
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: cc-by-nc-sa-4.0
|
| 3 |
+
base_model:
|
| 4 |
+
- utter-project/mHuBERT-147
|
| 5 |
+
---
|
| 6 |
+
|
| 7 |
+
This is a `mHuBERT-147` model traced into a pytorch jit file.
|
| 8 |
+
This allows to run and deploy the model without extra dependencies.
|
| 9 |
+
Model after 2nd training iteration is combined with faiss index, to output hubert labels.
|
| 10 |
+
|
| 11 |
+
##Usage
|
| 12 |
+
|
| 13 |
+
```
|
| 14 |
+
import torch
|
| 15 |
+
import soundfile as sf
|
| 16 |
+
|
| 17 |
+
extractor = torch.jit.load("mhubert_fp32_cuda.jit").to(torch.device("cuda"))
|
| 18 |
+
wav, sr = sf.read(
|
| 19 |
+
"rms_arctic_a0001.wav",
|
| 20 |
+
dtype="int16"
|
| 21 |
+
)
|
| 22 |
+
assert sr == 16000
|
| 23 |
+
x = torch.tensor(wav).unsqueeze(0).cuda()
|
| 24 |
+
x_len = torch.tensor([x.shape[1]]).cuda()
|
| 25 |
+
labels = traced_extractor(x, x_len)[1].detach().cpu().numpy()
|
| 26 |
+
```
|