Spaces:
Sleeping
Sleeping
Commit
·
bbd9e13
1
Parent(s):
52e3665
Fix missed file
Browse files- demo/delightful_univnet.py +74 -0
demo/delightful_univnet.py
ADDED
@@ -0,0 +1,74 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import tempfile
|
2 |
+
|
3 |
+
from gradio import Checkbox, Dropdown, Interface, Textbox
|
4 |
+
import soundfile as sf
|
5 |
+
import torch
|
6 |
+
from voicefixer import VoiceFixer
|
7 |
+
|
8 |
+
from models.delightful_univnet import DelightfulUnivnet
|
9 |
+
from training.datasets.hifi_libri_dataset import speakers_hifi_ids
|
10 |
+
|
11 |
+
from .config import speakers_delightful_22050
|
12 |
+
|
13 |
+
delightful_checkpoint_path = "epoch=5816-step=390418.ckpt"
|
14 |
+
|
15 |
+
device = torch.device("cpu")
|
16 |
+
|
17 |
+
delightfulunivnet_22050 = DelightfulUnivnet(
|
18 |
+
delightful_checkpoint_path=delightful_checkpoint_path,
|
19 |
+
).to(device)
|
20 |
+
|
21 |
+
voicefixer = VoiceFixer()
|
22 |
+
|
23 |
+
|
24 |
+
def generate_audio(text: str, speaker_name: str, fix_voice: bool):
|
25 |
+
speaker = torch.tensor(
|
26 |
+
[speakers_delightful_22050[speaker_name]],
|
27 |
+
device=device,
|
28 |
+
)
|
29 |
+
with torch.no_grad():
|
30 |
+
wav = delightfulunivnet_22050.forward(text, speaker)
|
31 |
+
wav = wav.squeeze().detach().cpu().numpy()
|
32 |
+
|
33 |
+
if fix_voice:
|
34 |
+
# Save the numpy array to a temporary wav file
|
35 |
+
with tempfile.NamedTemporaryFile(suffix=".wav", delete=True) as input_file:
|
36 |
+
# Write to the temp wav file
|
37 |
+
sf.write(input_file.name, wav, delightfulunivnet_22050.sampling_rate)
|
38 |
+
|
39 |
+
with tempfile.NamedTemporaryFile(suffix=".wav", delete=True) as output_file:
|
40 |
+
voicefixer.restore(
|
41 |
+
input=input_file.name, # low quality .wav/.flac file
|
42 |
+
output=output_file.name, # save file path
|
43 |
+
cuda=False, # GPU acceleration off
|
44 |
+
mode=0,
|
45 |
+
)
|
46 |
+
|
47 |
+
# Read the wav file back into a numpy array
|
48 |
+
wav_vf, sampling_rate = sf.read(output_file.name)
|
49 |
+
|
50 |
+
return sampling_rate, wav_vf
|
51 |
+
|
52 |
+
return delightfulunivnet_22050.sampling_rate, wav
|
53 |
+
|
54 |
+
|
55 |
+
interfaceDelightfulUnuvnet22050 = Interface(
|
56 |
+
generate_audio,
|
57 |
+
[
|
58 |
+
Textbox(
|
59 |
+
label="Text",
|
60 |
+
value="As the snake shook its head, a deafening shout behind Harry made both of them jump.",
|
61 |
+
),
|
62 |
+
Dropdown(
|
63 |
+
label="Speaker",
|
64 |
+
choices=list(speakers_delightful_22050.keys()),
|
65 |
+
value=speakers_hifi_ids[0],
|
66 |
+
),
|
67 |
+
Checkbox(
|
68 |
+
label="Fix voice (Voicefixer)",
|
69 |
+
value=False,
|
70 |
+
),
|
71 |
+
],
|
72 |
+
outputs="audio",
|
73 |
+
title=f"Delightful UnivNet, Sampling Rate: {delightfulunivnet_22050.sampling_rate}. When Voicefixer is enabled, the Simpling Rate is 44100.",
|
74 |
+
)
|