app.py

model = None
sid = ""

import io
import gradio as gr
import librosa
import numpy as np
import soundfile
from inference.infer_tool import Svc

import os
def list_files_tree(directory, indent=""):
    items = os.listdir(directory)
    for i, item in enumerate(items):
        prefix = "└── " if i == len(items) - 1 else "├── "
        print(indent + prefix + item)
        item_path = os.path.join(directory, item)
        if os.path.isdir(item_path):
            next_indent = indent + ("    " if i == len(items) - 1 else "│   ")
            list_files_tree(item_path, next_indent)

from huggingface_hub import snapshot_download
print("Models...")
models_id = """None1145/So-VITS-SVC-Vulpisfoglia"""
for model_id in models_id.split("\n"):
    if model_id in ["", " "]:
        break
    print(f"{model_id}...")
    snapshot_download(repo_id=model_id, local_dir=f"./Models/{model_id}")
    print(f"{model_id}!!!")
print("Models!!!")
list_files_tree("./")

import re
models_info = {}
models_folder_path = "./Models/None1145"
folder_names = [name for name in os.listdir(models_folder_path) if os.path.isdir(os.path.join(models_folder_path, name))]
for folder_name in folder_names:
    speaker = folder_name[12:]
    pattern = re.compile(r"G_(\d+)\.pth$")
    max_value = -1
    max_file = None
    models_path = f"{models_folder_path}/{folder_name}/Models"
    config_path = f"{models_folder_path}/{folder_name}/Configs"
    for filename in os.listdir(models_path):
        match = pattern.search(filename)
        if match:
            value = int(match.group(1))
            if value > max_value:
                max_value = value
                max_file = filename
    models_info[speaker] = {}
    models_info[speaker]["model"] = f"{models_path}/{max_file}"
    models_info[speaker]["config"] = f"{config_path}/config.json"
    if os.path.exists(f"{models_path}/feature_and_index.pkl"):
        models_info[speaker]["cluster"] = f"{models_path}/feature_and_index.pkl"
    elif os.path.exists(f"{models_path}/kmeans_10000.pt"):
        models_info[speaker]["cluster"] = f"{models_path}/kmeans_10000.pt"
    else:
        models_info[speaker]["cluster"] = ""
speakers = list(models_info.keys())

def load(speaker):
    global sid
    global model
    sid = speaker
    model = Svc(models_info[speaker]["model"], models_info[speaker]["config"], cluster_model_path=models_info[speaker]["cluster"])
    return "加载成功"
load(speakers[0])

def vc_fn(input_audio, vc_transform, auto_f0,cluster_ratio, slice_db, noise_scale):
    global sid
    if input_audio is None:
        return "You need to upload an audio", None
    sampling_rate, audio = input_audio
    # print(audio.shape,sampling_rate)
    duration = audio.shape[0] / sampling_rate
    # if duration > 90:
    #     return "请上传小于90s的音频，需要转换长音频请本地进行转换", None
    audio = (audio / np.iinfo(audio.dtype).max).astype(np.float32)
    if len(audio.shape) > 1:
        audio = librosa.to_mono(audio.transpose(1, 0))
    if sampling_rate != 16000:
        audio = librosa.resample(audio, orig_sr=sampling_rate, target_sr=16000)
    print(audio.shape)
    out_wav_path = "temp.wav"
    soundfile.write(out_wav_path, audio, 16000, format="wav")
    print( cluster_ratio, auto_f0, noise_scale)
    _audio = model.slice_inference(out_wav_path, sid, vc_transform, slice_db, cluster_ratio, auto_f0, noise_scale)
    return "Success", (44100, _audio)

app = gr.Blocks()
with app:
    with gr.Tabs():
        with gr.TabItem("Model"):
            speaker = gr.Dropdown(label="讲话人", choices=speakers, value=speakers[0])
            model_submit = gr.Button("加载模型", variant="primary")
            model_output1 = gr.Textbox(label="Output Message")
        model_submit.click(load, [speaker], [model_output1])
        with gr.TabItem("Basic"):
            # sid = gr.Dropdown(label="音色", choices=speakers, value=speakers[0])
            vc_input3 = gr.Audio(label="上传音频")
            vc_transform = gr.Number(label="变调（整数，可以正负，半音数量，升高八度就是12）", value=0)
            cluster_ratio = gr.Number(label="聚类模型混合比例，0-1之间，默认为0不启用聚类，能提升音色相似度，但会导致咬字下降（如果使用建议0.5左右）", value=0)
            auto_f0 = gr.Checkbox(label="自动f0预测，配合聚类模型f0预测效果更好,会导致变调功能失效（仅限转换语音，歌声不要勾选此项会究极跑调）", value=False)
            slice_db = gr.Number(label="切片阈值", value=-40)
            noise_scale = gr.Number(label="noise_scale 建议不要动，会影响音质，玄学参数", value=0.4)
            vc_submit = gr.Button("转换", variant="primary")
            vc_output1 = gr.Textbox(label="Output Message")
            vc_output2 = gr.Audio(label="Output Audio")
        vc_submit.click(vc_fn, [vc_input3, vc_transform,auto_f0,cluster_ratio, slice_db, noise_scale], [vc_output1, vc_output2])
    app.launch()