import os import glob import json import argparse import traceback import logging import gradio as gr import numpy as np import librosa import torch import asyncio import edge_tts from datetime import datetime from fairseq import checkpoint_utils from infer_pack.models import SynthesizerTrnMs256NSFsid, SynthesizerTrnMs256NSFsid_nono from vc_infer_pipeline import VC from config import Config config = Config() logging.getLogger("numba").setLevel(logging.WARNING) limitation = os.getenv("SYSTEM") == "spaces" # limit audio length in huggingface spaces def create_vc_fn(tgt_sr, net_g, vc, if_f0, file_index): def vc_fn( input_audio, f0_up_key, f0_method, index_rate, tts_mode, tts_text, tts_voice ): try: if tts_mode: if len(tts_text) > 100 and limitation: return "Text is too long", None if tts_text is None or tts_voice is None: return "You need to enter text and select a voice", None asyncio.run(edge_tts.Communicate(tts_text, "-".join(tts_voice.split('-')[:-1])).save("tts.mp3")) audio, sr = librosa.load("tts.mp3", sr=16000, mono=True) else: if input_audio is None: return "You need to upload an audio", None sampling_rate, audio = input_audio duration = audio.shape[0] / sampling_rate if duration > 20 and limitation: return "Please upload an audio file that is less than 20 seconds. If you need to generate a longer audio file, please use Colab.", 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) times = [0, 0, 0] f0_up_key = int(f0_up_key) audio_opt = vc.pipeline( hubert_model, net_g, 0, audio, times, f0_up_key, f0_method, file_index, index_rate, if_f0, f0_file=None, ) print( f"[{datetime.now().strftime('%Y-%m-%d %H:%M')}]: npy: {times[0]}, f0: {times[1]}s, infer: {times[2]}s" ) return "Success", (tgt_sr, audio_opt) except: info = traceback.format_exc() print(info) return info, (None, None) return vc_fn def load_hubert(): global hubert_model models, _, _ = checkpoint_utils.load_model_ensemble_and_task( ["hubert_base.pt"], suffix="", ) hubert_model = models[0] hubert_model = hubert_model.to(config.device) if config.is_half: hubert_model = hubert_model.half() else: hubert_model = hubert_model.float() hubert_model.eval() def change_to_tts_mode(tts_mode): if tts_mode: return gr.Audio.update(visible=False), gr.Textbox.update(visible=True), gr.Dropdown.update(visible=True) else: return gr.Audio.update(visible=True), gr.Textbox.update(visible=False), gr.Dropdown.update(visible=False) if __name__ == '__main__': load_hubert() categories = [] tts_voice_list = asyncio.get_event_loop().run_until_complete(edge_tts.list_voices()) voices = [f"{v['ShortName']}-{v['Gender']}" for v in tts_voice_list] with open("weights/folder_info.json", "r", encoding="utf-8") as f: folder_info = json.load(f) for category_name, category_info in folder_info.items(): if not category_info['enable']: continue category_title = category_info['title'] category_folder = category_info['folder_path'] description = category_info['description'] models = [] with open(f"weights/{category_folder}/model_info.json", "r", encoding="utf-8") as f: models_info = json.load(f) for model_name, info in models_info.items(): if not info['enable']: continue model_title = info['title'] model_author = info.get("author", None) model_cover = f"weights/{category_folder}/{model_name}/{info['cover']}" model_index = f"weights/{category_folder}/{model_name}/{info['feature_retrieval_library']}" cpt = torch.load(f"weights/{category_folder}/{model_name}/{model_name}.pth", map_location="cpu") tgt_sr = cpt["config"][-1] cpt["config"][-3] = cpt["weight"]["emb_g.weight"].shape[0] # n_spk if_f0 = cpt.get("f0", 1) if if_f0 == 1: net_g = SynthesizerTrnMs256NSFsid(*cpt["config"], is_half=config.is_half) else: net_g = SynthesizerTrnMs256NSFsid_nono(*cpt["config"]) del net_g.enc_q print(net_g.load_state_dict(cpt["weight"], strict=False)) net_g.eval().to(config.device) if config.is_half: net_g = net_g.half() else: net_g = net_g.float() vc = VC(tgt_sr, config) print(f"Model loaded: {model_name}") models.append((model_name, model_title, model_author, model_cover, create_vc_fn(tgt_sr, net_g, vc, if_f0, model_index))) categories.append([category_title, category_folder, description, models]) with gr.Blocks() as app: gr.Markdown( "#