app.py

import time
import matplotlib.pyplot as plt
import IPython.display as ipd
import re
import os
import json
import math
import torch
from torch import nn
from torch.nn import functional as F
from torch.utils.data import DataLoader
import gradio as gr
import commons
import utils
from data_utils import TextAudioLoader, TextAudioCollate, TextAudioSpeakerLoader, TextAudioSpeakerCollate
from models import SynthesizerTrn
from text.symbols import symbols
from text import text_to_sequence
import unicodedata
from scipy.io.wavfile import write
import openai

def get_text(text, hps):
    text_norm = text_to_sequence(text, hps.data.text_cleaners)
    if hps.data.add_blank:
        text_norm = commons.intersperse(text_norm, 0)
    text_norm = torch.LongTensor(text_norm)
    return text_norm

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

def selection(speaker):
    if speaker == "高咲侑（误）":
        spk = 0
        return spk

    elif speaker == "歩夢":
        spk = 1
        return spk

    elif speaker == "かすみ":
        spk = 2
        return spk

    elif speaker == "しずく":
        spk = 3
        return spk

    elif speaker == "果林":
        spk = 4
        return spk
    
    elif speaker == "愛":
        spk = 5
        return spk

    elif speaker == "彼方":
        spk = 6
        return spk

    elif speaker == "せつ菜":
        spk = 7
        return spk
    elif speaker == "エマ":
        spk = 8
        return spk
    elif speaker == "璃奈":
        spk = 9
        return spk
    elif speaker == "栞子":
        spk = 10
        return spk
    elif speaker == "ランジュ":
        spk = 11
        return spk
    elif speaker == "ミア":
        spk = 12
        return spk
    elif speaker == "三色绘恋1":
        spk = 13
        return spk
    elif speaker == "三色绘恋2":
        spk = 15
        return spk
    elif speaker == "派蒙":
        spk = 16
        return spk
def friend_chat(text,key,call_name,tts_input3):
  call_name = call_name
  openai.api_key = key
  identity = tts_input3
  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 is_japanese(string):
        for ch in string:
            if ord(ch) > 0x3040 and ord(ch) < 0x30FF:
                return True
        return False    
def sle(language,text,tts_input2,call_name,tts_input3):
    if language == "中文":
        tts_input1 = "[ZH]" + text.replace('\n','。').replace(' ',',') + "[ZH]"
        return tts_input1
    if language == "对话":
        text = friend_chat(text,tts_input2,call_name,tts_input3).replace('\n','。').replace(' ',',')
        text = f"[JA]{text}[JA]" if is_japanese(text) else f"[ZH]{text}[ZH]"
        return text
    elif language == "日文":
        tts_input1 = "[JA]" + text.replace('\n','。').replace(' ',',') + "[JA]"
        return tts_input1
def infer(text,tts_input2,tts_input3,language, speaker_id,n_scale= 0.667,n_scale_w = 0.8, l_scale = 1):
    speaker_name = speaker_id
    speaker_id = int(selection(speaker_id))
    text = sle(language,text,tts_input2,speaker_name,tts_input3)
    response = text.replace("[ZH]","").replace("[JA]","")
    stn_tst = get_text(text, hps_ms)
    with torch.no_grad():
        x_tst = stn_tst.unsqueeze(0).to(dev)
        x_tst_lengths = torch.LongTensor([stn_tst.size(0)]).to(dev)
        sid = torch.LongTensor([speaker_id]).to(dev)
        t1 = time.time()
        audio = net_g_ms.infer(x_tst, x_tst_lengths, sid=sid, noise_scale=n_scale, noise_scale_w=n_scale_w, length_scale=l_scale)[0][0,0].data.cpu().float().numpy()
        t2 = time.time()
        spending_time = "推理时间："+str(t2-t1)+"s" 
        print(spending_time)
    return response,(hps_ms.data.sampling_rate, audio) 
lan = ["中文","日文","对话"]
idols = ["高咲侑（误）","歩夢","かすみ","しずく","果林","愛","彼方","せつ菜","璃奈","栞子","エマ","ランジュ","ミア","派蒙"]


dev = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
hps_ms = utils.get_hparams_from_file("config.json")
net_g_ms = SynthesizerTrn(
    len(symbols),
    hps_ms.data.filter_length // 2 + 1,
    hps_ms.train.segment_size // hps_ms.data.hop_length,
    n_speakers=hps_ms.data.n_speakers,
    **hps_ms.model).to(dev)
_ = net_g_ms.eval()

_ = utils.load_checkpoint("G_1415000.pth", net_g_ms, None)

inputs = [gr.TextArea(label="输入你的文本，支持vits版在另一个仓库", value="一次審査、二次審査、それぞれの欄に記入をお願いします。"),
           gr.TextArea(label="如需使用openai，输入你的openai-key,合成方式选择对话", value="官网"),
           gr.TextArea(label="写上你给她的设定", value=""),
           gr.Dropdown(label="选择合成方式",choices=lan, value="日文", interactive=True),
           gr.Dropdown(label="选择说话人",choices=idols, value="かすみ", interactive=True),
           gr.Slider(minimum= 0.01,maximum=1.0,label="更改噪声比例", value=0.127),
           gr.Slider(minimum= 0.01,maximum=1.0,label="更改噪声偏差", value=0.8),
           gr.Slider(minimum= 0.1,maximum=10,label="更改时间比例", value=1)]
outputs=[ gr.Textbox(label="回复"),gr.Audio(type="numpy", label="Output audio")]
iface = gr.Interface(
    fn=infer,
    inputs=inputs,
    outputs=outputs,
    title="轻量化vits",
    description="虹团12人模型",
)
iface.launch()