File size: 6,813 Bytes
4de73fc 996feab 4de73fc 996feab 4de73fc 161b131 4de73fc 22980ff 996feab 4de73fc 161b131 4de73fc 9da3b76 4de73fc 161b131 4de73fc d94bbcb |
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 |
import logging
logging.getLogger('numba').setLevel(logging.WARNING)
logging.getLogger('matplotlib').setLevel(logging.WARNING)
logging.getLogger('urllib3').setLevel(logging.WARNING)
from text import text_to_sequence
import numpy as np
from scipy.io import wavfile
import torch
import json
import commons
import utils
import sys
import pathlib
import onnxruntime as ort
import gradio as gr
import argparse
import time
import os
from scipy.io.wavfile import write
def is_japanese(string):
for ch in string:
if ord(ch) > 0x3040 and ord(ch) < 0x30FF:
return True
return False
def is_english(string):
import re
pattern = re.compile('^[A-Za-z0-9.,:;!?()_*"\' ]+$')
if pattern.fullmatch(string):
return True
else:
return False
def to_numpy(tensor: torch.Tensor):
return tensor.detach().cpu().numpy() if tensor.requires_grad \
else tensor.detach().numpy()
def get_symbols_from_json(path):
assert os.path.isfile(path)
with open(path, 'r') as f:
data = json.load(f)
return data['symbols']
def sle(language,text):
text = text.replace('\n','。').replace(' ',',')
if language == "中文":
tts_input1 = "[ZH]" + text + "[ZH]"
return tts_input1
elif language == "自动":
tts_input1 = f"[JA]{text}[JA]" if is_japanese(text) else f"[ZH]{text}[ZH]"
return tts_input1
elif language == "日文":
tts_input1 = "[JA]" + text + "[JA]"
return tts_input1
elif language == "英文":
tts_input1 = "[EN]" + text + "[EN]"
return tts_input1
elif language == "手动":
return text
def get_text(text,hps_ms):
text_norm = text_to_sequence(text,hps_ms.symbols,hps_ms.data.text_cleaners)
if hps_ms.data.add_blank:
text_norm = commons.intersperse(text_norm, 0)
text_norm = torch.LongTensor(text_norm)
return text_norm
def create_tts_fn(ort_sess, speaker_id):
def tts_fn(text , language, n_scale= 0.667,n_scale_w = 0.8, l_scale = 1 ):
text =sle(language,text)
seq = text_to_sequence(text,hps.symbols, cleaner_names=hps.data.text_cleaners)
if hps.data.add_blank:
seq = commons.intersperse(seq, 0)
with torch.no_grad():
x = np.array([seq], dtype=np.int64)
x_len = np.array([x.shape[1]], dtype=np.int64)
sid = np.array([speaker_id], dtype=np.int64)
scales = np.array([n_scale, n_scale_w, l_scale], dtype=np.float32)
scales.resize(1, 3)
ort_inputs = {
'input': x,
'input_lengths': x_len,
'scales': scales,
'sid': sid
}
t1 = time.time()
audio = np.squeeze(ort_sess.run(None, ort_inputs))
audio *= 32767.0 / max(0.01, np.max(np.abs(audio))) * 0.6
audio = np.clip(audio, -32767.0, 32767.0)
t2 = time.time()
spending_time = "推理时间："+str(t2-t1)+"s"
print(spending_time)
return (hps.data.sampling_rate, audio)
return tts_fn
if __name__ == '__main__':
symbols = get_symbols_from_json('checkpoints/ShojoKageki/config.json')
hps = utils.get_hparams_from_file('checkpoints/ShojoKageki/config.json')
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
models = []
schools = ["ShojoKageki-Nijigasaki","ShojoKageki","Nijigasaki"]
lan = ["中文","日文","自动","手动"]
with open("checkpoints/info.json", "r", encoding="utf-8") as f:
models_info = json.load(f)
for i in models_info:
school = models_info[i]
speakers = school["speakers"]
checkpoint = school["checkpoint"]
phone_dict = {
symbol: i for i, symbol in enumerate(symbols)
}
ort_sess = ort.InferenceSession(checkpoint)
content = []
for j in speakers:
sid = int(speakers[j]['sid'])
title = school
example = speakers[j]['speech']
name = speakers[j]["name"]
content.append((sid, name, title, example, create_tts_fn(ort_sess, sid)))
models.append(content)
with gr.Blocks() as app:
gr.Markdown(
"# <center> vits-models\n"
)
with gr.Tabs():
for i in schools:
with gr.TabItem(i):
for (sid, name, title, example, tts_fn) in models[schools.index(i)]:
with gr.TabItem(name):
'''
with gr.Row():
gr.Markdown(
'<div align="center">'
f'<a><strong>{name}</strong></a>'
f'<img style="width:auto;height:300px;" src="file/{sid}.png">'
'</div>'
)
'''
with gr.Row():
with gr.Column():
with gr.Row():
with gr.Column():
gr.Markdown(
'<div align="center">'
f'<a><strong>{name}</strong></a>'
f'<img style="width:auto;height:400px;" src="file/image/{name}.png">'
'</div>'
)
input2 = gr.Dropdown(label="Language", choices=lan, value="自动", interactive=True)
with gr.Column():
input1 = gr.TextArea(label="Text", value=example)
input4 = gr.Slider(minimum=0, maximum=1.0, label="更改噪声比例(noise scale)，以控制情感", value=0.667)
input5 = gr.Slider(minimum=0, maximum=1.0, label="更改噪声偏差(noise scale w)，以控制音素长短", value=0.8)
input6 = gr.Slider(minimum=0.1, maximum=10, label="duration", value=1)
btnVC = gr.Button("Submit")
output1 = gr.Audio(label="采样率22050")
btnVC.click(tts_fn, inputs=[input1, input2, input4, input5, input6], outputs=[output1])
app.launch()
|