tts-generation-webui/tts_webui/rvc_tab/infer_rvc.py

# import os
# import sys
# from dotenv import load_dotenv
# from tts_webui.rvc_tab.hide_argv import hide_argv

# os.environ.setdefault("weight_root", "data/models/rvc/checkpoints")
# os.environ.setdefault("weight_uvr5_root", "data/models/rvc/uvr5_weights")
# os.environ.setdefault("index_root", "data/models/rvc/checkpoints")
# os.environ.setdefault("outside_index_root", "data/models/rvc/checkpoints")
# os.environ.setdefault("rmvpe_root", "data/models/rvc/rmvpe")


# import rvc_pkg

# rvc_dir = os.path.dirname(rvc_pkg.__file__)
# sys.path.append(rvc_dir)
# from rvc_pkg.configs.config import Config
# from rvc_pkg.infer.modules.vc.modules import VC

# sys.path.remove(rvc_dir)

# from tts_webui.rvc_tab.get_and_load_hubert import get_and_load_hubert_new, download_rmvpe


# last_model_path = None
# vc = None


# def infer_rvc(
#     input_path: str,  # Input path
#     index_path_2: str,  # Index path
#     model_name: str,  # Model name (stored in assets/weight_root)
#     device: str,  # Device
#     f0up_key: int = 0,  # F0up key
#     f0method: str = "harvest",  # F0 method (harvest or pm)
#     index_rate: float = 0.66,  # Index rate
#     is_half: bool = False,  # Use half -> True
#     filter_radius: int = 3,  # Filter radius
#     resample_sr: int = 0,  # Resample sample rate
#     rms_mix_rate: float = 1,  # RMS mix rate
#     protect: float = 0.33,  # Protect breath sounds
# ):
#     global last_model_path
#     load_dotenv()
#     with hide_argv():
#         config = Config()
#     config.device = device if device else config.device
#     config.is_half = is_half if is_half else config.is_half
#     global vc
#     if vc is None:
#         vc = VC(config)
#         if vc.hubert_model is None:
#             vc.hubert_model = get_and_load_hubert_new(config)
#     if last_model_path != model_name:
#         vc.get_vc(model_name)
#         last_model_path = model_name
#     if f0method == "rmvpe":
#         download_rmvpe()
#     message, wav_opt = vc.vc_single(
#         0,
#         input_path,
#         f0up_key,
#         None,
#         f0method,
#         None,
#         os.path.join("data\\models\\rvc\\checkpoints\\", index_path_2),
#         index_rate,
#         filter_radius,
#         resample_sr,
#         rms_mix_rate,
#         protect,
#     )
#     print(message)
#     return wav_opt, {
#         "original_audio_path": input_path,
#         "index_path": index_path_2,
#         "model_path": model_name,
#         "f0method": f0method,
#         "f0up_key": f0up_key,
#         "index_rate": index_rate,
#         "device": device,
#         "is_half": is_half,
#         "filter_radius": filter_radius,
#         "resample_sr": resample_sr,
#         "rms_mix_rate": rms_mix_rate,
#         "protect": protect,
#     }


# if __name__ == "__main__":
#     rate, wav_data = infer_rvc(
#         input_path="sample.wav",
#         model_name="Voltina.pth",
#         index_path_2="Voltina.index",
#         device="cuda:0",
#         f0up_key=0,
#         f0method="harvest",
#         index_rate=0.66,
#         is_half=False,
#         filter_radius=3,
#         resample_sr=0,
#         rms_mix_rate=1,
#         protect=0.33,
#     )

#     from scipy.io.wavfile import write

#     write("out.wav", rate, wav_data)