app.py

import gradio as gr
import numpy as np
import torch
import os
import re_matching
from tools.sentence import split_by_language, sentence_split
import utils
from infer import infer, latest_version, get_net_g
import gradio as gr
import webbrowser
from config import config
from tools.translate import translate
from tools.webui import reload_javascript

device = config.webui_config.device
if device == "mps":
    os.environ["PYTORCH_ENABLE_MPS_FALLBACK"] = "1"


def speak_fn(
        text: str,
        exceed_flag,
        speaker="TalkFlower_CNzh",
        sdp_ratio=0.2,      # SDP/DP混合比
        noise_scale=0.6,        # 感情
        noise_scale_w=0.6,      # 音素长度
        length_scale=0.9,       # 语速
        language="ZH",
        interval_between_para=0.2,      # 段间间隔
        interval_between_sent=1,        # 句间间隔
    ):
    while text.find("\n\n") != -1:
        text = text.replace("\n\n", "\n")
    if len(text) > 100:
        print(f"Too Long Text: {text}")
        gr.Warning("Too long! No more than 100 characters. 一口气不要超过 100 个字，憋坏我了。")
        if exceed_flag:
            return gr.update(value="./assets/audios/nomorethan100.wav", autoplay=True), False
        else:
            return gr.update(value="./assets/audios/overlength.wav", autoplay=True), True
    audio_list = []
    if len(text) > 42:
        print(f"Long Text: {text}")
        para_list = re_matching.cut_para(text)
        for p in para_list:
            audio_list_sent = []
            sent_list = re_matching.cut_sent(p)
            for s in sent_list:
                audio = infer(
                    s,
                    sdp_ratio=sdp_ratio,
                    noise_scale=noise_scale,
                    noise_scale_w=noise_scale_w,
                    length_scale=length_scale,
                    sid=speaker,
                    language=language,
                    hps=hps,
                    net_g=net_g,
                    device=device,
                )
                audio_list_sent.append(audio)
                silence = np.zeros((int)(44100 * interval_between_sent))
                audio_list_sent.append(silence)
            if (interval_between_para - interval_between_sent) > 0:
                silence = np.zeros(
                    (int)(44100 * (interval_between_para - interval_between_sent))
                )
                audio_list_sent.append(silence)
            audio16bit = gr.processing_utils.convert_to_16_bit_wav(
                np.concatenate(audio_list_sent)
            )  # 对完整句子做音量归一
            audio_list.append(audio16bit)
    else:
        print(f"Short Text: {text}")
        silence = np.zeros(hps.data.sampling_rate // 2, dtype=np.int16)
        with torch.no_grad():
            for piece in text.split("|"):
                audio = infer(
                    piece,
                    sdp_ratio=sdp_ratio,
                    noise_scale=noise_scale,
                    noise_scale_w=noise_scale_w,
                    length_scale=length_scale,
                    sid=speaker,
                    language=language,
                    hps=hps,
                    net_g=net_g,
                    device=device,
                )
                audio16bit = gr.processing_utils.convert_to_16_bit_wav(audio)
                audio_list.append(audio16bit)
                audio_list.append(silence)  # 将静音添加到列表中
    
    audio_concat = np.concatenate(audio_list)    
    return (hps.data.sampling_rate, audio_concat), exceed_flag


def init_fn():
    gr.Info("2023-11-23: 花钱买了稍微好一点的服务器，现在生成应该更快了。")
    gr.Info("2023-11-24: 优化长句生成效果；增加示例；更新了一些小彩蛋。")
    gr.Info("Only support Chinese now. Trying to train a mutilingual model.")


with open("./css/style.css", "r", encoding="utf-8") as f:
    customCSS = f.read()

with gr.Blocks(css=customCSS) as demo:
    exceed_flag = gr.State(value=False)
    talkingFlowerPic = gr.HTML("""<img src="file=assets/flower-2x.webp" alt="TalkingFlowerPic">""", elem_id="talking_flower_pic")
    input_text = gr.Textbox(lines=1, label="Talking Flower will say:", elem_classes="wonder-card", elem_id="input_text")
    speak_button = gr.Button("Speak!", elem_id="speak_button", elem_classes="button wonder-card")
    gr.Examples(["你今天好棒", "雄蕊羊痒的", "我一朵花好害怕", "再来找我玩哦", "冲呀冲呀！", "塔塔开！一字摸塔塔开！", "好了是闺蜜，不好嘞是敌咪"], label=None, inputs=[input_text], elem_id="examples")
    audio_output = gr.Audio(label="输出音频", show_label=False, autoplay=True, elem_id="audio_output", elem_classes="wonder-card")
    
    
    demo.load(
        init_fn,
        inputs=[],
        outputs=[]
    )
    input_text.submit(
        speak_fn,
        inputs=[input_text, exceed_flag],
        outputs=[audio_output, exceed_flag],
    )
    speak_button.click(
        speak_fn,
        inputs=[input_text, exceed_flag],
        outputs=[audio_output, exceed_flag],
    )


if __name__ == "__main__":
    hps = utils.get_hparams_from_file(config.webui_config.config_path)
    version = hps.version if hasattr(hps, "version") else latest_version
    net_g = get_net_g(model_path=config.webui_config.model, version=version, device=device, hps=hps)
    reload_javascript()
    demo.launch(
        allowed_paths=["./assets"],
        show_api=False,
        # server_name=server_name,
        # server_port=server_port,
        inbrowser=True,
    )