-
Notifications
You must be signed in to change notification settings - Fork 0
/
run_GUI_demo_shan.py
226 lines (208 loc) · 7.81 KB
/
run_GUI_demo_shan.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
import os
import gradio as gr
import torch
import torch.cuda
from Utility.utils import float2pcm
from Architectures.ControllabilityGAN.GAN import GanWrapper
from InferenceInterfaces.ToucanTTSInterface import ToucanTTSInterface
from Utility.storage_config import MODELS_DIR
from Utility.utils import load_json_from_path
demo = gr.Blocks()
class ControllableInterface:
def __init__(self, gpu_id="cpu", available_artificial_voices=1000):
if gpu_id == "cpu":
os.environ["CUDA_VISIBLE_DEVICES"] = ""
else:
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"] = f"{gpu_id}"
self.device = "cuda" if torch.cuda.is_available() else "cpu"
self.model = ToucanTTSInterface(device=self.device, tts_model_path="Shan")
self.wgan = GanWrapper(
os.path.join(MODELS_DIR, "Embedding", "embedding_gan.pt"),
device=self.device,
)
self.generated_speaker_embeds = list()
self.available_artificial_voices = available_artificial_voices
self.current_language = ""
self.current_accent = ""
def read(
self,
prompt,
language,
accent,
voice_seed,
duration_scaling_factor,
pause_duration_scaling_factor,
pitch_variance_scale,
energy_variance_scale,
emb_slider_1,
emb_slider_2,
emb_slider_3,
emb_slider_4,
emb_slider_5,
emb_slider_6,
):
if self.current_language != language:
self.model.set_phonemizer_language(language)
self.current_language = language
if self.current_accent != accent:
self.model.set_accent_language(accent)
self.current_accent = accent
self.wgan.set_latent(voice_seed)
controllability_vector = torch.tensor(
[
emb_slider_1,
emb_slider_2,
emb_slider_3,
emb_slider_4,
emb_slider_5,
emb_slider_6,
],
dtype=torch.float32,
)
embedding = self.wgan.modify_embed(controllability_vector)
self.model.set_utterance_embedding(embedding=embedding)
phones = self.model.text2phone.get_phone_string(prompt)
if len(phones) > 1800:
prompt = "Your input was too long. Please try either a shorter text or split it into several parts."
if self.current_language != "eng":
self.model.set_phonemizer_language("eng")
self.current_language = "eng"
if self.current_accent != "eng":
self.model.set_accent_language("eng")
self.current_accent = "eng"
print(prompt)
wav, sr, fig = self.model(
prompt,
input_is_phones=False,
duration_scaling_factor=duration_scaling_factor,
pitch_variance_scale=pitch_variance_scale,
energy_variance_scale=energy_variance_scale,
pause_duration_scaling_factor=pause_duration_scaling_factor,
return_plot_as_filepath=True,
)
return sr, wav, fig
class TTSWebUI:
def __init__(
self,
gpu_id="cpu",
title="Controllable Text-to-Speech for over 7000 Languages",
article="",
available_artificial_voices=1000,
path_to_iso_list="Preprocessing/multilinguality/iso_to_fullname.json",
):
iso_to_name = load_json_from_path(path_to_iso_list)
text_selection = [
f"{iso_to_name[iso_code]} Text ({iso_code})" for iso_code in iso_to_name
]
# accent_selection = [f"{iso_to_name[iso_code]} Accent ({iso_code})" for iso_code in iso_to_name]
self.controllable_ui = ControllableInterface(
gpu_id=gpu_id, available_artificial_voices=available_artificial_voices
)
self.iface = gr.Interface(
fn=self.read,
inputs=[
gr.Textbox(
lines=2,
placeholder="write what you want the synthesis to read here...",
value="မႂ်ႇသုင်ၶႃႈ ယူႇလီၵိၼ်ဝၢၼ် ၵတ်းယဵၼ်ၸႂ် မိူၼ်ၾႃႉၾူၼ်လူမ်းလီယူႇၶႃႈ ၼေႃႈ",
label="Text input",
),
gr.Dropdown(
text_selection,
type="value",
value="Shan Text (shn)",
label="Select the Language of the Text (type on your keyboard to find it quickly)",
),
gr.Slider(
minimum=0,
maximum=available_artificial_voices,
step=1,
value=1000,
label="Random Seed for the artificial Voice",
),
gr.Slider(
minimum=0.7,
maximum=1.3,
step=0.1,
value=1.2,
label="Duration Scale",
),
gr.Slider(
minimum=0.5,
maximum=1.5,
step=0.1,
value=1.0,
label="Pitch Variance Scale",
),
gr.Slider(
minimum=0.5,
maximum=1.5,
step=0.1,
value=1.0,
label="Energy Variance Scale",
),
gr.Slider(
minimum=-10.0,
maximum=10.0,
step=0.1,
value=10.0,
label="Femininity / Masculinity",
),
gr.Slider(
minimum=-10.0,
maximum=10.0,
step=0.1,
value=-10.0,
label="Voice Depth",
),
],
outputs=[
gr.Audio(type="numpy", label="Speech"),
gr.Image(label="Visualization"),
],
title=title,
theme="default",
allow_flagging="never",
article=article,
)
def read(
self,
prompt,
language,
voice_seed,
duration_scaling_factor,
pitch_variance_scale,
energy_variance_scale,
emb1,
emb2,
):
sr, wav, fig = self.controllable_ui.read(
prompt=prompt,
language=language.split(" ")[-1].split("(")[1].split(")")[0],
accent=language.split(" ")[-1].split("(")[1].split(")")[0],
voice_seed=voice_seed,
duration_scaling_factor=duration_scaling_factor,
pause_duration_scaling_factor=1.0,
pitch_variance_scale=pitch_variance_scale,
energy_variance_scale=energy_variance_scale,
emb_slider_1=emb1,
emb_slider_2=emb2,
emb_slider_3=0.0,
emb_slider_4=0.0,
emb_slider_5=0.0,
emb_slider_6=0.0,
)
return (sr, float2pcm(wav)), fig
def render(self):
return self.iface
if __name__ == "__main__":
with gr.Blocks() as demo:
gr.Markdown(
"<p align='center' style='font-size: 20px;'><a href='https://github.com/DigitalPhonetics/IMS-Toucan'>IMS-Toucan</a>: Multilingual and Controllable Text-to-Speech Toolkit of the Speech and Language Technologies Group at the University of Stuttgart.</p>"
)
gr.HTML(
"<p align='center' style='font-size: 18px;'><a href='https://github.com/NoerNova/IMS-Toucan-Shan'>IMS-Toucan-Shan</a>: Contain the Shan finetune script</p>"
)
TTSWebUI(gpu_id="cuda" if torch.cuda.is_available() else "cpu").render()
demo.launch()