custom_model.py

from transformers import PreTrainedModel, PretrainedConfig, AutoConfig, AutoModel
from spleeter.separator import Separator


class SpleeterConfig(PretrainedConfig):
    model_type = "spleeter"

    def __init__(self, stems=2, **kwargs):
        super().__init__(**kwargs)
        self.stems = stems


class SpleeterModel(PreTrainedModel):
    config_class = SpleeterConfig

    def __init__(self, config):
        super().__init__(config)
        self.separator = Separator(f"spleeter:{config.stems}stems")

    def forward(self, audio_path: str):
        """
        Separates the stems in the given audio file.
        Args:
            audio_path (str): Path to the input audio file.
        Returns:
            path: Separated stems.
        """
        return self.separator.separate_to_file(audio_path, "separated_audio")


AutoConfig.register("spleeter", SpleeterConfig)
AutoModel.register(SpleeterConfig, SpleeterModel)
SpleeterConfig.register_for_auto_class()
SpleeterModel.register_for_auto_class("AutoModel")