New feature: OnnxPredict algorithm

[xaviliz]

New feature: OnnxPredict algorithm

Feature

It makes additional changes in Essentia library to build ONNX Runtime Inferencing library from the source and to implement a new algorithm OnnxPredict for running ONNX models (.onnx) with multiple IO.

Implementation

• Provide a new building script for ONNX Runtime Inferencing library.
• Modify Essentia scripts to link with the onnxruntime dynamic library.
• Implement new algorithm OnnxPredict to run ONNX models in Essentia.
• Implement unittests in test_onnxpredict.py

Prerequisites

• python >= 3.10
• cmake >= 3.28

Testing

• Builds successfully with ONNX Runtime v1.22.1 in MacOS
  • ARM64
  • x86_64
• Builds successfully with ONNX Runtime v1.22.1 in Linux
• Multiple input inferencing
• Multiple output inferencing
• No runtime errors or compatibility issues

How to Test

Tested in onnxruntime-v1.22.1:

• MacOS with an ARM64 machine with python 3.13.4 and cmake 4.0.2
• Linux docker with python 3.10.18 and cmake 4.1.0

How to build ONNX Runtime

After installing Essentia dependencies in a virtual environment, install cmake

python3 -m pip install cmake
which cmake

Then we can run the building script:

cd packaging/debian_3rdparty
bash build_onnx.sh

How to build OnnxPredict

In MacOS:

source .env/bin/activate
python3 waf configure --fft=KISS --include-algos=OnnxPredict,Windowing,Spectrum,MelBands,UnaryOperator,TriangularBands,FFT,Magnitude,NoiseAdder,RealAccumulator,FileOutputProxy,FrameCutter --static-dependencies --pkg-config-path=/packaging/debian_3rdparty/lib/pkgconfig --with-onnx --lightweight= --with-python --pythondir=.env/lib/python3.13/site-packages
python3 waf -v && python3 waf install

In Linux:

python3 waf configure --fft=KISS --include-algos=OnnxPredict,Windowing,Spectrum,MelBands,UnaryOperator,TriangularBands,FFT,Magnitude,NoiseAdder,RealAccumulator,FileOutputProxy,FrameCutter --static-dependencies --with-onnx --lightweight= --with-python --pkg-config-path /usr/share/pkgconfig --std=c++14
python3 waf -v && python3 waf install

How to unittest

# prepare essentia audio repo
git clone https://github.com/MTG/essentia-audio.git test/essentia-audio
rm -rf test/audio && mv test/essentia-audio test/audio

# download effnet.onnx model for testing
curl https://essentia.upf.edu/models/feature-extractors/discogs-effnet/discogs-effnet-bsdynamic-1.onnx --output test/models/discogs-effnet-bsdynamic-1.onnx
python3 test/src/unittests/all_tests.py onnxpredict

[xaviliz]

Thank you @palonso, all changes/suggestions have been addressed.
Please when you can revise and give me feedback for the changes.
I think it is almost done.

Before merging it would be nice to:

• Test the algorithm in Linux with the last changes, but it should work fine.
• Test it in OSX with METAL and OPEN_ML.
• Make a brew cask to test the essentia installation with the onnxruntime package
  • I think it was failing because it was builded without some static dependencies flag.