Spaces:
Runtime error
Runtime error
Upload app.py
Browse files
app.py
ADDED
@@ -0,0 +1,108 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from models import SynthesizerTrn
|
2 |
+
from vits_pinyin import VITS_PinYin
|
3 |
+
from text import cleaned_text_to_sequence
|
4 |
+
from text.symbols import symbols
|
5 |
+
import gradio as gr
|
6 |
+
import utils
|
7 |
+
import torch
|
8 |
+
import argparse
|
9 |
+
import os
|
10 |
+
import re
|
11 |
+
import logging
|
12 |
+
|
13 |
+
logging.getLogger('numba').setLevel(logging.WARNING)
|
14 |
+
limitation = os.getenv("SYSTEM") == "spaces"
|
15 |
+
|
16 |
+
|
17 |
+
def create_calback(net_g: SynthesizerTrn, tts_front: VITS_PinYin):
|
18 |
+
def tts_calback(text, dur_scale):
|
19 |
+
if limitation:
|
20 |
+
text_len = len(re.sub("\[([A-Z]{2})\]", "", text))
|
21 |
+
max_len = 150
|
22 |
+
if text_len > max_len:
|
23 |
+
return "Error: Text is too long", None
|
24 |
+
|
25 |
+
phonemes, char_embeds = tts_front.chinese_to_phonemes(text)
|
26 |
+
input_ids = cleaned_text_to_sequence(phonemes)
|
27 |
+
with torch.no_grad():
|
28 |
+
x_tst = torch.LongTensor(input_ids).unsqueeze(0).to(device)
|
29 |
+
x_tst_lengths = torch.LongTensor([len(input_ids)]).to(device)
|
30 |
+
x_tst_prosody = torch.FloatTensor(
|
31 |
+
char_embeds).unsqueeze(0).to(device)
|
32 |
+
audio = net_g.infer(x_tst, x_tst_lengths, x_tst_prosody, noise_scale=0.5,
|
33 |
+
length_scale=dur_scale)[0][0, 0].data.cpu().float().numpy()
|
34 |
+
del x_tst, x_tst_lengths, x_tst_prosody
|
35 |
+
return "Success", (16000, audio)
|
36 |
+
|
37 |
+
return tts_calback
|
38 |
+
|
39 |
+
|
40 |
+
example = [['天空呈现的透心的蓝,像极了当年。总在这样的时候,透过窗棂,心,在天空里无尽的游弋!柔柔的,浓浓的,痴痴的风,牵引起心底灵动的思潮;情愫悠悠,思情绵绵,风里默坐,红尘中的浅醉,诗词中的优柔,任那自在飞花轻似梦的情怀,裁一束霓衣,织就清浅淡薄的安寂。', 1],
|
41 |
+
['风的影子翻阅过淡蓝色的信笺,柔和的文字浅浅地漫过我安静的眸,一如几朵悠闲的云儿,忽而氤氲成汽,忽而修饰成花,铅华洗尽后的透彻和靓丽,爽爽朗朗,轻轻盈盈', 1],
|
42 |
+
['时光仿佛有穿越到了从前,在你诗情画意的眼波中,在你舒适浪漫的暇思里,我如风中的思绪徜徉广阔天际,仿佛一片沾染了快乐的羽毛,在云环影绕颤动里浸润着风的呼吸,风的诗韵,那清新的耳语,那婉约的甜蜜,那恬淡的温馨,将一腔情澜染得愈发的缠绵。', 1],]
|
43 |
+
|
44 |
+
|
45 |
+
if __name__ == "__main__":
|
46 |
+
parser = argparse.ArgumentParser()
|
47 |
+
parser.add_argument("--share", action="store_true",
|
48 |
+
default=False, help="share gradio app")
|
49 |
+
args = parser.parse_args()
|
50 |
+
|
51 |
+
device = torch.device("cpu")
|
52 |
+
|
53 |
+
# pinyin
|
54 |
+
tts_front = VITS_PinYin("./bert", device)
|
55 |
+
|
56 |
+
# config
|
57 |
+
hps = utils.get_hparams_from_file("./configs/bert_vits.json")
|
58 |
+
|
59 |
+
# model
|
60 |
+
net_g = SynthesizerTrn(
|
61 |
+
len(symbols),
|
62 |
+
hps.data.filter_length // 2 + 1,
|
63 |
+
hps.train.segment_size // hps.data.hop_length,
|
64 |
+
**hps.model)
|
65 |
+
|
66 |
+
model_path = "vits_bert_model.pth"
|
67 |
+
utils.load_model(model_path, net_g)
|
68 |
+
net_g.eval()
|
69 |
+
net_g.to(device)
|
70 |
+
|
71 |
+
tts_calback = create_calback(net_g, tts_front)
|
72 |
+
|
73 |
+
app = gr.Blocks()
|
74 |
+
with app:
|
75 |
+
gr.Markdown("# Best TTS based on BERT and VITS with some Natural Speech Features Of Microsoft\n\n"
|
76 |
+
"code : github.com/PlayVoice/vits_chinese\n\n"
|
77 |
+
"1, Hidden prosody embedding from BERT,get natural pauses in grammar\n\n"
|
78 |
+
"2, Infer loss from NaturalSpeech,get less sound error\n\n"
|
79 |
+
"3, Framework of VITS,get high audio quality\n\n"
|
80 |
+
"<video id='video' controls='' preload='yes'>\n\n"
|
81 |
+
"<source id='mp4' src='https://user-images.githubusercontent.com/16432329/220678182-4775dec8-9229-4578-870f-2eebc3a5d660.mp4' type='video/mp4'>\n\n"
|
82 |
+
"</videos>\n\n"
|
83 |
+
)
|
84 |
+
|
85 |
+
with gr.Tabs():
|
86 |
+
with gr.TabItem("TTS"):
|
87 |
+
with gr.Row():
|
88 |
+
with gr.Column():
|
89 |
+
textbox = gr.TextArea(label="Text",
|
90 |
+
placeholder="Type your sentence here (Maximum 150 words)",
|
91 |
+
value="中文语音合成", elem_id=f"tts-input")
|
92 |
+
duration_slider = gr.Slider(minimum=0.1, maximum=5, value=1, step=0.1,
|
93 |
+
label='速度 Speed')
|
94 |
+
with gr.Column():
|
95 |
+
text_output = gr.Textbox(label="Message")
|
96 |
+
audio_output = gr.Audio(
|
97 |
+
label="Output Audio", elem_id="tts-audio")
|
98 |
+
btn = gr.Button("Generate!")
|
99 |
+
btn.click(tts_calback,
|
100 |
+
inputs=[textbox, duration_slider],
|
101 |
+
outputs=[text_output, audio_output])
|
102 |
+
gr.Examples(
|
103 |
+
examples=example,
|
104 |
+
inputs=[textbox, duration_slider],
|
105 |
+
outputs=[text_output, audio_output],
|
106 |
+
fn=tts_calback
|
107 |
+
)
|
108 |
+
app.queue(concurrency_count=3).launch(show_api=False, share=args.share)
|