File size: 11,763 Bytes
5c5159a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d15d7a0
5c5159a
 
 
 
 
 
 
 
 
 
 
 
 
d15d7a0
 
 
5c5159a
d15d7a0
5c5159a
 
 
 
 
 
 
d15d7a0
53c4934
d15d7a0
5c5159a
 
 
 
 
 
 
 
 
 
 
 
 
 
d15d7a0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5c5159a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
aa89e52
 
 
 
5c5159a
 
 
 
 
 
 
 
 
d15d7a0
 
 
 
 
5c5159a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d15d7a0
5c5159a
 
 
d15d7a0
5c5159a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d15d7a0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5c5159a
d15d7a0
 
 
 
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
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
import re
import gradio as gr
import torch
import unicodedata
import commons
import utils
import pathlib
from models import SynthesizerTrn
from text import text_to_sequence
import time
import os
import io
from scipy.io.wavfile import write
from flask import Flask, request
from threading import Thread
import openai
import requests
import json
import soundfile as sf
from scipy import signal
class VitsGradio:
    def __init__(self):
        self.lan = ["中文","日文","自动"]
        self.chatapi = ["gpt-3.5-turbo","gpt3"]
        self.modelPaths = []
        for root,dirs,files in os.walk("checkpoints"):
            for dir in dirs:
                self.modelPaths.append(dir)
        with gr.Blocks() as self.Vits:
            with gr.Tab("调试用"):
                with gr.Row():
                    with gr.Column():
                        with gr.Row():
                            with gr.Column():
                                self.text = gr.TextArea(label="Text", value="你好")
                                with gr.Accordion(label="测试api", open=False):
                                    self.local_chat1 = gr.Checkbox(value=False, label="使用网址+文本进行模拟")
                                    self.url_input = gr.TextArea(label="键入测试", value="http://127.0.0.1:8080/chat?Text=")
                                    butto = gr.Button("模拟前端抓取语音文件")
                                btnVC = gr.Button("测试tts+对话程序")
                            with gr.Column():
                                output2 = gr.TextArea(label="回复")
                                output1 = gr.Audio(label="采样率22050")
                                output3 = gr.outputs.File(label="44100hz: output.wav")
                butto.click(self.Simul, inputs=[self.text, self.url_input], outputs=[output2,output3])
                btnVC.click(self.tts_fn, inputs=[self.text], outputs=[output1,output2])
            with gr.Tab("控制面板"):
                with gr.Row():
                    with gr.Column():
                        with gr.Row():
                            with gr.Column():
                                self.api_input1 = gr.TextArea(label="输入gpt/茉莉云的api-key或本地存储说话模型的路径.如果要用茉莉云则用'|'隔开key和密码", value="49eig5nu3rllvg6e|itcn9760")
                                with gr.Accordion(label="chatbot选择(默认gpt3.5)", open=False):
                                                self.api_input2 = gr.Checkbox(value=False, label="茉莉云")
                                                self.local_chat1 = gr.Checkbox(value=False, label="启动本地chatbot")
                                                self.local_chat2 = gr.Checkbox(value=False, label="是否量化")
                                                res = gr.TextArea()
                                Botselection = gr.Button("完成chatbot设定")
                                Botselection.click(self.check_bot, inputs=[self.api_input1,self.api_input2,self.local_chat1,self.local_chat2], outputs = [res])
                                self.input1 = gr.Dropdown(label = "模型", choices = self.modelPaths, value = self.modelPaths[0], type = "value")
                                self.input2 = gr.Dropdown(label="Language", choices=self.lan, value="自动", interactive=True)
                            with gr.Column():
                                btnVC = gr.Button("完成vits TTS端设定")
                                self.input3 = gr.Dropdown(label="Speaker", choices=list(range(1001)), value=0, interactive=True)
                                self.input4 = gr.Slider(minimum=0, maximum=1.0, label="更改噪声比例(noise scale),以控制情感", value=0.6)
                                self.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=0.667)
                                self.input6 = gr.Slider(minimum=0.1, maximum=10, label="duration", value=1)
                                statusa = gr.TextArea()
                btnVC.click(self.create_tts_fn, inputs=[self.input1, self.input2, self.input3, self.input4, self.input5, self.input6], outputs = [statusa])

    def Simul(self,text,url_input):
        web = url_input + text
        res = requests.get(web)
        music = res.content
        with open('output.wav', 'wb') as code:
            code.write(music)
        file_path = "output.wav"
        return web,file_path


    def mori(self,text):
        import http.client
        conn = http.client.HTTPSConnection("api.mlyai.com")
        payload = json.dumps({
            "content": text,
            "type": 1,
            "from": "123456",
            "fromName": "侑"
        })
        headers = {
            'Api-Key': self.api_key,
            'Api-Secret': self.api_secret,
            'Content-Type': 'application/json'
        }
        conn.request("POST", "/reply", payload, headers)
        res = conn.getresponse()
        data = res.read()
        decoded_data = json.loads(data.decode("utf-8")) 

        if decoded_data["code"] == "00000":
            answer = decoded_data["data"][0]["content"]
            if text == 'exit':
                conn.close()
            return answer
        else:
            conn.close()
            return '对不起,做不到'

    def chatgpt(self,text):
        self.messages.append({"role": "user", "content": text},)
        chat = openai.ChatCompletion.create(model="gpt-3.5-turbo", messages= self.messages)
        reply = chat.choices[0].message.content
        return reply
    
    def ChATGLM(self,text):
        if text == 'clear':
            self.history = []
        response, new_history = self.model.chat(self.tokenizer, text, self.history)
        response = response.replace(" ",'').replace("\n",'.')
        self.history = new_history
        return response
    
    def gpt3_chat(self,text):
        call_name = "Waifu"
        openai.api_key = args.key
        identity = ""
        start_sequence = '\n'+str(call_name)+':'
        restart_sequence = "\nYou: "
        if 1 == 1:
            prompt0 = text #当期prompt
        if text == 'quit':
            return prompt0
        prompt = identity + prompt0 + start_sequence
        response = openai.Completion.create(
            model="text-davinci-003",
            prompt=prompt,
            temperature=0.5,
            max_tokens=1000,
            top_p=1.0,
            frequency_penalty=0.5,
            presence_penalty=0.0,
            stop=["\nYou:"]
        )
        return response['choices'][0]['text'].strip()
    
    def check_bot(self,api_input1,api_input2,local_chat1,local_chat2):
        try:
            self.api_key, self.api_secret = api_input1.split("|")
        except:
            pass
        if local_chat1:
            from transformers import AutoTokenizer, AutoModel
            self.tokenizer = AutoTokenizer.from_pretrained(api_input1, trust_remote_code=True)
            if local_chat2:
                self.model = AutoModel.from_pretrained(api_input1, trust_remote_code=True).half().quantize(4).cuda()
            else:
                self.model = AutoModel.from_pretrained(api_input1, trust_remote_code=True)
            self.history = []
        else:
            try:
                self.messages = []
                openai.api_key = api_input1
            except:
                pass
        return "Finished"
    
    def is_japanese(self,string):
        for ch in string:
            if ord(ch) > 0x3040 and ord(ch) < 0x30FF:
                return True
        return False
    
    def is_english(self,string):
        import re
        pattern = re.compile('^[A-Za-z0-9.,:;!?()_*"\' ]+$')
        if pattern.fullmatch(string):
            return True
        else:
            return False

    

    def get_text(self,text, hps, cleaned=False):
        if cleaned:
            text_norm = text_to_sequence(text, self.hps_ms.symbols, [])
        else:
            text_norm = text_to_sequence(text, self.hps_ms.symbols, self.hps_ms.data.text_cleaners)
        if self.hps_ms.data.add_blank:
            text_norm = commons.intersperse(text_norm, 0)
        text_norm = torch.LongTensor(text_norm)
        return text_norm


    def get_label(self,text, label):
        if f'[{label}]' in text:
            return True, text.replace(f'[{label}]', '')
        else:
            return False, text

    def sle(self,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 self.is_japanese(text) else f"[ZH]{text}[ZH]"
            return tts_input1
        elif language == "日文":
            tts_input1 = "[JA]" + text + "[JA]"
            return tts_input1

    def create_tts_fn(self,path, input2, input3, n_scale= 0.667,n_scale_w = 0.8, l_scale = 1 ):
        self.language = input2
        self.speaker_id = int(input3)
        self.n_scale = n_scale
        self.n_scale_w = n_scale_w
        self.l_scale = l_scale
        self.dev = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
        self.hps_ms = utils.get_hparams_from_file(f"checkpoints/{path}/config.json")
        self.n_speakers = self.hps_ms.data.n_speakers if 'n_speakers' in self.hps_ms.data.keys() else 0
        self.n_symbols = len(self.hps_ms.symbols) if 'symbols' in self.hps_ms.keys() else 0
        self.net_g_ms = SynthesizerTrn(
            self.n_symbols,
            self.hps_ms.data.filter_length // 2 + 1,
            self.hps_ms.train.segment_size // self.hps_ms.data.hop_length,
            n_speakers=self.n_speakers,
            **self.hps_ms.model).to(self.dev)
        _ = self.net_g_ms.eval()
        _ = utils.load_checkpoint(f"checkpoints/{path}/model.pth", self.net_g_ms)
        return 'success'


    def tts_fn(self,text):
        if self.local_chat1:
            text = self.mori(text)
        elif self.api_input2:
            text = self.ChATGLM(text)
        else:
            text = text = self.chatgpt(text)
        print(text)
        text =self.sle(self.language,text)
        with torch.no_grad():
            stn_tst = self.get_text(text, self.hps_ms, cleaned=False)
            x_tst = stn_tst.unsqueeze(0).to(self.dev)
            x_tst_lengths = torch.LongTensor([stn_tst.size(0)]).to(self.dev)
            sid = torch.LongTensor([self.speaker_id]).to(self.dev)
            audio = self.net_g_ms.infer(x_tst, x_tst_lengths, sid=sid, noise_scale=self.n_scale, noise_scale_w=self.n_scale_w, length_scale=self.l_scale)[0][
                0, 0].data.cpu().float().numpy()
        resampled_audio_data = signal.resample(audio, len(audio) * 2)
        sf.write('temp.wav', resampled_audio_data, 44100, 'PCM_24')
        return (self.hps_ms.data.sampling_rate, audio),text.replace('[JA]','').replace('[ZH]','')

app = Flask(__name__)
print("开始部署")
grVits = VitsGradio()

@app.route('/chat')
def text_api():
    message = request.args.get('Text','')
    audio,text = grVits.tts_fn(message)
    text = text.replace('[JA]','').replace('[ZH]','')
    with open('temp.wav','rb') as bit:
        wav_bytes = bit.read()
    headers = {
            'Content-Type': 'audio/wav',
            'Text': text.encode('utf-8')}
    return wav_bytes, 200, headers

def gradio_interface():
    return grVits.Vits.launch()

if __name__ == '__main__':
    api_thread = Thread(target=app.run, args=("0.0.0.0", 8080))
    gradio_thread = Thread(target=gradio_interface)
    api_thread.start()
    gradio_thread.start()