- Load audio files into google drive folder and update path in babbling_cnn.ipynb.
- Run babbling_cnn.ipynb until header titled Stop. This should create 2 npy files.
- Move npy files to folder containing run_model.py and run_model2.py on cluster.
- Run run_model.py and run_model2.py to train models and get testing results. Note: requires numpy, keras, pandas, and matplotlib.
- babbling_cnn.ipynb:
- extracts spectrograms from audio files
- saves spectrogram arrays to npy files, one for French and one for English
- includes code to load spectrograms from npy files, initialize two models, and train models. this code will not run due to colab memory limitations.
- feature_test.ipynb:
- initial shared notebook from project
- contains original code to extract spectrograms & initialize models
- contains code to extract specialized features
- run_model.py:
- load spectrograms from npy files
- initalize, train, and test model 1 (CNN only)
- save accuracy & loss plots
- run_model2.py:
- load spectrograms from npy files,
- initalize, train, and test model 2 (CNN with BLSTM)
- save accuracy & loss plots
- data_english.npy and data_french.npy: generated by babbling_cnn.ipynb. Not included here (too large for github and lfs.)