Summary: Easy to use vocal separation on CLI or as a python package, using the amazing MDX-Net models from UVR trained by @Anjok07
Audio Separator is a Python package that allows you to separate an audio file into two stems, primary and secondary, using a model in the ONNX format trained by @Anjok07 for use with UVR (https://github.com/Anjok07/ultimatevocalremovergui).
The primary stem typically contains the instrumental part of the audio, while the secondary stem contains the vocals, but in some models this is reversed.
- Separate audio into instrumental and vocal stems.
- Supports all common audio formats (WAV, MP3, FLAC, M4A, etc.)
- Ability to specify a pre-trained deep learning model in ONNX format.
- CLI support for easy use in scripts and batch processing.
- Python API for integration into other projects.
๐ฌ If successfully configured, you should see this log message when running audio-separator:
ONNXruntime has CUDAExecutionProvider available, enabling acceleration
Conda: conda install pytorch=*=*cuda* onnxruntime=*=*cuda* audio-separator -c pytorch -c conda-forge
Pip: pip install "audio-separator[gpu]"
Docker: beveradb/audio-separator:gpu
Colab has recently upgraded to CUDA 12, which isn't yet supported by the official ONNX Runtime releases.
To get audio-separator
working with GPU acceleration on colab, first install this onnxruntime-gpu
wheel (which I built with CUDA 12 support):
pip install https://github.com/karaokenerds/python-audio-separator/releases/download/v0.12.1/onnxruntime_gpu-1.17.0-cp310-cp310-linux_x86_64.whl
Then install audio-separator:
pip install "audio-separator[gpu]"
Conda: conda install audio-separator-c pytorch -c conda-forge
Pip: pip install "audio-separator[cpu]"
Docker: beveradb/audio-separator
๐ฌ If successfully configured, you should see this log message when running audio-separator:
ONNXruntime has CoreMLExecutionProvider available, enabling acceleration
Pip: pip install "audio-separator[silicon]"
If you installed audio-separator
using pip
, you'll separately need to ensure you have ffmpeg
installed.
This should be easy to install on most platforms, e.g.:
๐ง Debian/Ubuntu: apt-get update; apt-get install -y ffmpeg
๏ฃฟ macOS:brew update; brew install ffmpeg
In theory, all you should need to do to get audio-separator
working with a GPU is install it with the [gpu]
extra as above.
However, sometimes getting both PyTorch and ONNX Runtime working with CUDA support can be a bit tricky so it may not work that easily.
You may need to reinstall both packages directly, allowing pip to calculate the right versions for your platform:
pip uninstall torch onnxruntime
pip cache purge
pip install --force-reinstall torch torchvision torchaudio
pip install --force-reinstall onnxruntime-gpu
Depending on your hardware, you may get better performance with the optimum version of onnxruntime:
pip install --force-reinstall "optimum[onnxruntime-gpu]"
Depending on your CUDA version and hardware, you may need to install torch from the cu118
index instead:
pip install --force-reinstall torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
Note: if anyone knows how to make this cleaner so we can support both different platform-specific dependencies for hardware acceleration without a separate installation process for each, please let me know or raise a PR!
There are images published on Docker Hub for GPU (CUDA) and CPU inferencing, for both amd64
and arm64
platforms.
You probably want to volume-mount a folder containing whatever file you want to separate, which can then also be used as the output folder.
For example, if the current directory contains your input file input.wav
, you could run audio-separator
like so:
docker run -it -v `pwd`:/workdir beveradb/audio-separator input.wav
If you're using a machine with a GPU, you'll want to use the GPU specific image and pass in the GPU device to the container, like this:
docker run -it --gpus all -v `pwd`:/workdir beveradb/audio-separator:gpu input.wav
If the GPU isn't being detected, make sure your docker runtime environment is passing through the GPU correctly - there are various guides online to help with that.
You can use Audio Separator via the command line:
usage: audio-separator [-h] [-v] [--log_level LOG_LEVEL] [--model_name MODEL_NAME] [--model_file_dir MODEL_FILE_DIR] [--output_dir OUTPUT_DIR] [--output_format OUTPUT_FORMAT] [--denoise DENOISE] [--normalize NORMALIZE]
[--single_stem SINGLE_STEM] [--invert_spect INVERT_SPECT] [--samplerate SAMPLERATE] [--adjust ADJUST] [--dim_c DIM_C] [--hop HOP] [--segment_size SEGMENT_SIZE] [--overlap overlap] [--batch_size BATCH_SIZE]
[audio_file]
Separate audio file into different stems.
positional arguments:
audio_file The audio file path to separate, in any common format.
options:
-h, --help show this help message and exit
-v, --version show program's version number and exit
--log_level LOG_LEVEL Optional: logging level, e.g. info, debug, warning (default: info). Example: --log_level=debug
--model_name MODEL_NAME Optional: model name to be used for separation (default: UVR_MDXNET_KARA_2). Example: --model_name=UVR-MDX-NET-Inst_HQ_3
--model_file_dir MODEL_FILE_DIR Optional: model files directory (default: /tmp/audio-separator-models/). Example: --model_file_dir=/app/models
--output_dir OUTPUT_DIR Optional: directory to write output files (default: <current dir>). Example: --output_dir=/app/separated
--output_format OUTPUT_FORMAT Optional: output format for separated files, any common format (default: FLAC). Example: --output_format=MP3
--denoise DENOISE Optional: enable or disable denoising during separation (default: True). Example: --denoise=False
--normalize NORMALIZE Optional: enable or disable normalization during separation (default: True). Example: --normalize=False
--single_stem SINGLE_STEM Optional: output only single stem, either instrumental or vocals. Example: --single_stem=instrumental
--invert_spect INVERT_SPECT Optional: invert secondary stem using spectogram (default: True). Example: --invert_spect=False
--samplerate SAMPLERATE Optional: samplerate (default: 44100). Example: --samplerate=44100
--adjust ADJUST Optional: adjust (default: 1). Example: --adjust=1
--dim_c DIM_C Optional: dim_c (default: 4). Example: --dim_c=4
--hop HOP Optional: hop (default: 1024). Example: --hop=1024
--segment_size SEGMENT_SIZE Optional: segment_size (default: 256). Example: --segment_size=256
--overlap overlap Optional: overlap (default: 0.25). Example: --overlap=0.25
--batch_size BATCH_SIZE Optional: batch_size (default: 4). Example: --batch_size=4
Example:
audio-separator /path/to/your/audio.wav --model_name UVR_MDXNET_KARA_2
This command will process the file and generate two new files in the current directory, one for each stem.
You can use Audio Separator in your own Python project. Here's how you can use it:
from audio_separator.separator import Separator
# Initialize the Separator class (with optional configuration properties below)
separator = Separator()
# Load a machine learning model (if unspecified, defaults to 'UVR-MDX-NET-Inst_HQ_3')
separator.load_model()
# Perform the separation on specific audio files without reloading the model
primary_stem_path, secondary_stem_path = separator.separate('audio1.wav')
print(f'Primary stem saved at {primary_stem_path}')
print(f'Secondary stem saved at {secondary_stem_path}')
You can process multiple separations without reloading the model, to save time and memory.
You only need to load a model when choosing or changing models. See example below:
from audio_separator.separator import Separator
# Initialize the Separator with other configuration properties below
separator = Separator()
# Load a model
separator.load_model('UVR-MDX-NET-Inst_HQ_3')
# Separate multiple audio files without reloading the model
output_file_paths_1 = separator.separate('audio1.wav')
output_file_paths_2 = separator.separate('audio2.wav')
output_file_paths_3 = separator.separate('audio3.wav')
# Load a different model
separator.load_model('UVR_MDXNET_KARA_2')
# Separate the same files with the new model
output_file_paths_4 = separator.separate('audio1.wav')
output_file_paths_5 = separator.separate('audio2.wav')
output_file_paths_6 = separator.separate('audio3.wav')
- audio_file: The path to the audio file to be separated. Supports all common formats (WAV, MP3, FLAC, M4A, etc.)
- log_level: (Optional) Logging level, e.g. info, debug, warning. Default: INFO
- log_formatter: (Optional) The log format. Default: '%(asctime)s - %(name)s - %(levelname)s - %(message)s'
- model_name: (Optional) The name of the model to use for separation. Defaults to 'UVR-MDX-NET-Inst_HQ_3', a very powerful model for Karaoke instrumental tracks.
- model_file_dir: (Optional) Directory to cache model files in. Default: /tmp/audio-separator-models/
- output_dir: (Optional) Directory where the separated files will be saved. If not specified, outputs to current dir.
- output_format: (Optional) Format to encode output files, any common format (WAV, MP3, FLAC, M4A, etc.). Default: WAV
- denoise_enabled: (Optional) Flag to enable or disable denoising as part of the separation process. Default: True
- normalization_enabled: (Optional) Flag to enable or disable normalization as part of the separation process. Default: False
- output_single_stem: (Optional) Output only single stem, either instrumental or vocals.
- invert_secondary_stem_using_spectogram=True,
- samplerate: (Optional) Modify the sample rate of the output audio. Default: 44100
- hop_length: (Optional) Hop length; advanced parameter used by the separation process. Default: 1024
- segment_size: (Optional) Segment size; advanced parameter used by the separation process. Default: 256
- overlap: (Optional) Overlap; advanced parameter used by the separation process. Default: 0.25
- batch_size: (Optional) Batch Size; advanced parameter used by the separation process. Default: 4
Python >= 3.9
Libraries: onnx, onnxruntime, numpy, librosa, torch, wget, six
This project uses Poetry for dependency management and packaging. Follow these steps to setup a local development environment:
- Make sure you have Python 3.9 or newer installed on your machine.
- Install Poetry by following the installation guide here.
Clone the repository to your local machine:
git clone https://github.com/YOUR_USERNAME/audio-separator.git
cd audio-separator
Replace YOUR_USERNAME with your GitHub username if you've forked the repository, or use the main repository URL if you have the permissions.
Run the following command to install the project dependencies:
poetry install
To activate the virtual environment, use the following command:
poetry shell
You can run the CLI command directly within the virtual environment. For example:
audio-separator path/to/your/audio-file.wav
Once you are done with your development work, you can exit the virtual environment by simply typing:
exit
To build the package for distribution, use the following command:
poetry build
This will generate the distribution packages in the dist directory - but for now only @beveradb will be able to publish to PyPI.
Contributions are very much welcome! Please fork the repository and submit a pull request with your changes, and I'll try to review, merge and publish promptly!
- This project is 100% open-source and free for anyone to use and modify as they wish.
- If the maintenance workload for this repo somehow becomes too much for me I'll ask for volunteers to share maintainership of the repo, though I don't think that is very likely
- Development and support for the MDX-Net separation models is part of the main UVR project, this repo is just a CLI/Python package wrapper to simplify running those models programmatically. So, if you want to try and improve the actual models, please get involved in the UVR project and look for guidance there!
This project is licensed under the MIT License.
- Please Note: If you choose to integrate this project into some other project using the default model or any other model trained as part of the UVR project, please honor the MIT license by providing credit to UVR and its developers!
- Anjok07 - Author of Ultimate Vocal Remover GUI, which almost all of the code in this repo was copied from! Definitely deserving of credit for anything good from this project. Thank you!
- DilanBoskan - Your contributions at the start of this project were essential to the success of UVR. Thank you!
- Kuielab & Woosung Choi - Developed the original MDX-Net AI code.
- KimberleyJSN - Advised and aided the implementation of the training scripts for MDX-Net and Demucs. Thank you!
- Hv - Helped implement chunks into the MDX-Net AI code. Thank you!
For questions or feedback, please raise an issue or reach out to @beveradb (Andrew Beveridge) directly.