Classify Applauses

Classify applauses and cheering apart from classical music.

The model trained here would be embedded in a client side application which filters applauses at runtime.

Installation

Use Anaconda.

$ conda install tensorflow-gpu 

Then run the Jupyter notebooks.

Data

Not included in this repository. It is possible to easily substitute with customized data. Contact the author for data.

Notebooks

• Preprocessing: converts .wav into .tfrecord
• Spectrogram: performs STFT on the sequential audio data to obtain 2 dimensional spectrograms
• Training: trains a CNN on the spectrograms
• Testing: evaluates precision/recall on test set and tests model on noisy data not in the training set

Model

• model: last Keras checkpoint
• classify_applauses_model.h5: Keras saved model (if saved correctly, same as above)
• tfjs_model: TF Layers model exported for tfjs use