Requires these dependencies (I'd recommend making a virtualenv -- see below):
- tensorflow -> https://www.tensorflow.org/install/
- python 3+ (python dependencies -> tensorflow, numpy, portaudio / pyaudio)
npm install - install npm dependencies
npm run build:model It downloads all the data files from AWS's S3 and trains the model.
Once the model is trained, you'll need to port your favorite checkpoint file to deeplearnjs using this command npm run build:portmodel This will run a bash script that prompts you to choose the model (amongst many possible checkpoints from previous steps) you want to port.
I ran in a virtual python environment with 'virtualenv' using source ~/tensorflow/bin/activate (because that's where I put the virtual environment)
npm run start - boot application
- take out parts of the violin signal that are not actually playing or too quiet to register...
- will the web audio smoothing help?
- still memory leaking?
- need a way to take out silent parts of training data besides audacity...
- create a 'loading model' spinner and load weights on aws if not local
- adjust screen size can somehow update width of state object??
- pitchfinder types... / faster/better algorithm
- better system of points (not all in one flattened object...?)