The program generates a model of feed-forward NN with 30 inputs and 13824 outputs.
30 inputs: FFT analyses of sounds
13824 outputs: sets the brightness for 13824 LEDs of the lightobject "Interspace #3"
Find a documentation about this work on: https://vimeo.com/280350114
The training data can be downloaded here and opened with liveTraining.py: www.birkschmithuesen.com/SAI/traingsdata.txt
A trained model is here and opened with loadModel.py: www.birkschmithuesen.com/SAI/model.h5
The program predicts the output for the lightobject in real time from received FFT data.
The communication is done via OSC.
WARNING: if you have no network card with the fixed IP 2.0.0.1 in your computer, the program will crash.
Also see: line 153 and line 102
Ableton Live(sound program)/FFT analysis => 30 float values via OSC (NN input) => python/neural network => 13824 float values via OSC (NN output) => JAVA(visualizer on screen and light object)
- TensorFlow (pip tensorflow)
- Keras (pip keras)
- PythonOsc (pip python-osc) www.birkschmithuesen.com/SAI/traingsdata.txt
- brew install portaudio
- pip3 install -r requirements.txt
- python=3.6.8
- create Conda Environment with libraries defined in "specs-conda.txt": conda create --name $ENV_name --file specs-conda.txt
- update tensorflow with pip according to "cpecs-pip.txt": pip install tensorflow==1.14