)))
https://github.com/marl/crepe
into the tuning app:
https://github.com/pterodactylus42/ccgt/tree/ml-beta
Planning to use the implementation org.tensorflow:tensorflow-lite:2.2.0 like in
https://github.com/projectmatris/antimalwareapp
according to
https://github.com/marl/crepe/issues/79
only the tiny model works in realtime android application.
Converted the models to .tflite by tweaking the crepe scripts. How to get there:
Begin the tutorial at
https://github.com/Hvass-Labs/TensorFlow-Tutorials/blob/master/01_Simple_Linear_Model.ipynb
In order to be able to do that, you have to install a working python environment which is more or less trivial depending on your system
On Mac, install a basic virtual environment using Conda:
https://docs.conda.io/projects/conda/en/latest/user-guide/tasks/manage-environments.html
Up next, install all missing packages in order to be able to follow the above-mentioned tutorial. Use this as a nutshell to start from:
conda create -n cenv python=3.9.2
conda activate cenv
conda install matplotlib
...
When it comes to installing crepe, jump into your environment and follow the instructions on the crepe GitHub page.
crepe's core.py was tweaked as follows:
...
import tensorflow as tf
...
package_dir = os.path.dirname(os.path.realpath(_file_))
filename = "model-{}.h5".format(model_capacity)
model.load_weights(os.path.join(package_dir, filename))
model.compile('adam', 'binary_crossentropy')
#spit out the converted module
converter = tf.lite.TFLiteConverter.from_keras_model(model)
tflmodel = converter.convert()
file = open('tflite_model.tflite' , 'wb')
file.write( tflmodel )
print('Model size ')
print(model_capacity)
print(' written as tflite model :-)')
#done
models[model_capacity] = model
...
