[Discussion] Spleeter works natively on Apple Silicon Macs if you use Apple's version of TensorFlow

Spleeter is working for me natively on my M1 Max MacBook Pro, built against Apple’s Metal version of TensorFlow.

Instructions from memory:

1. brew install miniforge
2. conda init "$(basename "${SHELL}")"
3. conda install -c apple tensorflow-deps numpy scipy llvmlite librosa pandas
4. pip install tensorflow-macos tensorflow-metal
5. git clone https://github.com/Deezer/spleeter && cd spleeter
6. pip install poetry
7. open the pyproject.toml and get rid of tensorflow/numpy/scipy/llvmlite/librosa/pandas
8. poetry update
9. poetry install
10. poetry build
11. pip install ./dist/spleeter-2.3.0-py3-none-any.whl

You might have to adapt some of these instructions, but it should broadly set you on the right path.

[AlexanderPavlenko]

Got it running too. MacBook Air (M1, 2020), macOS 12.2

# https://developer.apple.com/metal/tensorflow-plugin/

conda install -c apple tensorflow-deps==2.5.0
python -m pip install tensorflow-macos==2.5.0
python -m pip install tensorflow-metal
conda install -c apple llvmlite==0.36.0 numba pandas==1.3.3

git clone --depth 1 --branch v2.3.0 https://github.com/deezer/spleeter.git
# edit pyproject.toml :
# -tensorflow = "2.5.0"
# +tensorflow-macos = "2.5.0"
python -m pip install ./spleeter

brew install libsndfile
sudo ln -s /opt/homebrew/lib/libsndfile.dylib /usr/local/lib/

[romi1502]

Hi @argsnd, that’s great news thanks for this solution! we also managed to make it work on M1 but haven’t communicated yet. We’ll update the documentation very soon probably taking inspiration of this.