Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -7,21 +7,46 @@ import os
|
|
7 |
|
8 |
token=os.environ.get("key_")
|
9 |
tokenizer = AutoTokenizer.from_pretrained("wasmdashai/vtk",token=token)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
10 |
|
11 |
-
model=VitsModel.from_pretrained("wasmdashai/vtk",token=token).cuda()
|
12 |
zero = torch.Tensor([0]).cuda()
|
13 |
print(zero.device) # <-- 'cpu' 🤔
|
14 |
import torch
|
15 |
@spaces.GPU
|
16 |
-
def modelspeech(text):
|
17 |
|
18 |
|
19 |
inputs = tokenizer(text, return_tensors="pt")
|
|
|
20 |
with torch.no_grad():
|
21 |
wav = model(input_ids=inputs["input_ids"].cuda()).waveform.cpu().numpy().reshape(-1)#.detach()
|
22 |
|
23 |
return model.config.sampling_rate,wav#remove_noise_nr(wav)
|
24 |
|
25 |
-
|
26 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
27 |
demo.launch()
|
|
|
7 |
|
8 |
token=os.environ.get("key_")
|
9 |
tokenizer = AutoTokenizer.from_pretrained("wasmdashai/vtk",token=token)
|
10 |
+
models= {}
|
11 |
+
|
12 |
+
def get_model(name_model):
|
13 |
+
global models
|
14 |
+
if name_model in models:
|
15 |
+
return models[name_model]
|
16 |
+
models[name_model]=VitsModel.from_pretrained(name_model,token=token).cuda()
|
17 |
+
models[name_model].decoder.apply_weight_norm()
|
18 |
+
# torch.nn.utils.weight_norm(self.decoder.conv_pre)
|
19 |
+
# torch.nn.utils.weight_norm(self.decoder.conv_post)
|
20 |
+
for flow in models[name_model].flow.flows:
|
21 |
+
torch.nn.utils.weight_norm(flow.conv_pre)
|
22 |
+
torch.nn.utils.weight_norm(flow.conv_post)
|
23 |
+
return models[name_model]
|
24 |
+
|
25 |
|
|
|
26 |
zero = torch.Tensor([0]).cuda()
|
27 |
print(zero.device) # <-- 'cpu' 🤔
|
28 |
import torch
|
29 |
@spaces.GPU
|
30 |
+
def modelspeech(text,name_model):
|
31 |
|
32 |
|
33 |
inputs = tokenizer(text, return_tensors="pt")
|
34 |
+
model=get_model(name_model)
|
35 |
with torch.no_grad():
|
36 |
wav = model(input_ids=inputs["input_ids"].cuda()).waveform.cpu().numpy().reshape(-1)#.detach()
|
37 |
|
38 |
return model.config.sampling_rate,wav#remove_noise_nr(wav)
|
39 |
|
40 |
+
model_choices = gr.Dropdown(
|
41 |
+
choices=[
|
42 |
+
"wasmdashai/vits-ar-sa",
|
43 |
+
"wasmdashai/vits-ar-sa-huba",
|
44 |
+
"wasmdashai/vits-ar-sa-ms",
|
45 |
+
"wasmdashai/vits-ar-sa-magd",
|
46 |
+
"wasmdashai/vtk",
|
47 |
+
],
|
48 |
+
label="اختر النموذج",
|
49 |
+
value="wasmdashai/vtk",
|
50 |
+
)
|
51 |
+
demo = gr.Interface(fn=modelspeech, inputs=["text",model_choices], outputs=["audio"])
|
52 |
demo.launch()
|