app.py

import sys, os
import torch
import argparse
import commons
import utils
from models import SynthesizerTrn
from text.symbols import symbols
from text import cleaned_text_to_sequence, get_bert
from text.cleaner import clean_text
import gradio as gr
import webbrowser
import soundfile as sf
from datetime import datetime
import pytz


net_g = None
models = {
    "AdorableDarling": "./MODELS/adorabledarling.pth",
    "Silverleg": "./MODELS/J8900.pth",
    "Rrabbitt": "./MODELS/rabbit4900.pth",
    "MistyNikki": "./MODELS/nikki9400.pth",
    "LightHammer": "./MODELS/hammer.pth",
    "VivaciousViolet": "./MODELS/vv.pth",
    "AlluWin": "./MODELS/AW.pth",
    "Mainlade": "./MODELS/DLM.pth",
}

def get_text(text, language_str, hps):
    norm_text, phone, tone, word2ph = clean_text(text, language_str)
    phone, tone, language = cleaned_text_to_sequence(phone, tone, language_str)

    if hps.data.add_blank:
        phone = commons.intersperse(phone, 0)
        tone = commons.intersperse(tone, 0)
        language = commons.intersperse(language, 0)
        for i in range(len(word2ph)):
            word2ph[i] = word2ph[i] * 2
        word2ph[0] += 1
    bert = get_bert(norm_text, word2ph, language_str)
    del word2ph

    assert bert.shape[-1] == len(phone)

    phone = torch.LongTensor(phone)
    tone = torch.LongTensor(tone)
    language = torch.LongTensor(language)

    return bert, phone, tone, language

def infer(text, sdp_ratio, noise_scale, noise_scale_w, length_scale, sid, model_dir):
    global net_g
    bert, phones, tones, lang_ids = get_text(text, "ZH", hps)
    with torch.no_grad():
        x_tst=phones.to(device).unsqueeze(0)
        tones=tones.to(device).unsqueeze(0)
        lang_ids=lang_ids.to(device).unsqueeze(0)
        bert = bert.to(device).unsqueeze(0)
        x_tst_lengths = torch.LongTensor([phones.size(0)]).to(device)
        del phones
        speakers = torch.LongTensor([hps.data.spk2id[sid]]).to(device)
        audio = net_g.infer(x_tst, x_tst_lengths, speakers, tones, lang_ids, bert, sdp_ratio=sdp_ratio
                           , noise_scale=noise_scale, noise_scale_w=noise_scale_w, length_scale=length_scale)[0][0,0].data.cpu().float().numpy()
        del x_tst, tones, lang_ids, bert, x_tst_lengths, speakers
        sf.write("tmp.wav", audio, 44100)
        return audio

def convert_wav_to_mp3(wav_file):
    tz = pytz.timezone('Asia/Shanghai')
    now = datetime.now(tz).strftime('%m%d%H%M%S')
    os.makedirs('out', exist_ok=True)  
    output_path_mp3 = os.path.join('out', f"{now}.mp3")

    renamed_input_path = os.path.join('in', f"in.wav")
    os.makedirs('in', exist_ok=True)
    os.rename(wav_file.name, renamed_input_path)
    command = ["ffmpeg", "-i", renamed_input_path, "-acodec", "libmp3lame", "-y", output_path_mp3]
    os.system(" ".join(command))
    return output_path_mp3
    
def tts_generator(text, speaker, sdp_ratio, noise_scale, noise_scale_w, length_scale, model):
    global net_g
    model_path = models[model]
    net_g, _, _, _ = utils.load_checkpoint(model_path, net_g, None, skip_optimizer=True)
    try:
        with torch.no_grad():
            audio = infer(text, sdp_ratio=sdp_ratio, noise_scale=noise_scale, noise_scale_w=noise_scale_w, length_scale=length_scale, sid=speaker,model_dir=model)
        with open('tmp.wav', 'rb') as wav_file:
            mp3 = convert_wav_to_mp3(wav_file)  
        return "生成语音成功", (hps.data.sampling_rate, audio), mp3
    except Exception as e:
        return "生成语音失败：" + str(e), None, None


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

    speaker_ids = hps.data.spk2id
    speakers = list(speaker_ids.keys())

    with gr.Blocks() as app:
        with gr.Row():
            with gr.Column():

                gr.Markdown("测试用")
                text = gr.TextArea(label="Text", placeholder="Input Text Here",
                                value="在不在？能不能借给我三百块钱买可乐",
                                info="使用huggingface的免费CPU进行推理，因此速度不快，一次性不要输入超过500汉字")
                model = gr.Radio(choices=list(models.keys()), value=list(models.keys())[0], label='音声模型')
                #model = gr.Dropdown(choices=models,value=models[0], label='音声模型')
                speaker = gr.Radio(choices=speakers, value=speakers[0], label='Speaker')
                gr.Markdown(value="生成参数，效果玄学")
                sdp_ratio = gr.Slider(minimum=0, maximum=1, value=0.2, step=0.01, label='语调变化')
                noise_scale = gr.Slider(minimum=0.1, maximum=1.5, value=0.5, step=0.01, label='感情变化')
                noise_scale_w = gr.Slider(minimum=0.1, maximum=1.4, value=0.9, step=0.01, label='音节长度')
                length_scale = gr.Slider(minimum=0.1, maximum=2, value=1, step=0.01, label='生成语音总长度')
                btn = gr.Button("生成", variant="primary")
            with gr.Column():
                text_output = gr.Textbox(label="Message")
                audio_output = gr.Audio(label="试听")
                MP3_output = gr.File(label="下载")
                gr.Markdown("""
                
                """)
        btn.click(
                tts_generator,
                inputs=[text, speaker, sdp_ratio, noise_scale, noise_scale_w, length_scale, model],
                outputs=[text_output, audio_output,MP3_output]
                )

    app.launch(show_error=True)