BS-Roformer MLX

This folder contains the original Python implementation of BS-Roformer using MLX.

Setup

Ensure you have the required dependencies installed (refer to the root project or install manually):

pip install mlx numpy scipy pyyaml torch torchaudio

Usage

You can run the inference script python3 run_mlx.py. (CLI) or You can run the inference script python3 gui.py. (GUI)

From the project root (recommended for PYTHONPATH):

export PYTHONPATH=$PYTHONPATH:$(pwd)
python3 -m bs_roformer_python_legacy.run_mlx \
  --model_type v1_fast \
  --audio_path "/path/to/input.mp3" \
  --output_path "output_folder"

Options:

• --model_type: Model variant (default: v1_fast).
• --audio_path: Path to the input audio file.
• --output_path: Directory where separated stems will be saved.
• --start_offset: Start time in seconds (optional).
• --duration: Duration to process in seconds (optional).

Structure

• run_mlx.py: Main inference script.
• models/: Symlinked to the root models directory containing weights and config.