How can I write a Python program, that recognizes clapping sounds in an audio file, and saves them to disk?

Question

I would like to write a Python program that takes an audio file as input, recognizes clapping sounds in it, then extracts these sounds into a file.

My idea is the following:

1. Get (or create) a dataset containing many audio clips with clapping sounds in them,
2. Train a machine learning model to recognize these sounds,
3. Test the model on new audio files.

How can I achieve point 2? I already have a dataset ready, so I just need to run a ML algorithm on them. I've seen several alternatives:

1. Keras
2. PyTorch
3. Scikit-learn

But I have no idea which to choose for this particular task and how I would go about writing a Python script that does the training and testing. Can someone please help?

Answer 1

Claps have specific waves that could be detected with several solutions, including ML.

Here is a code using keras:

https://www.kaggle.com/code/carlolepelaars/bidirectional-lstm-for-audio-labeling-with-keras

In addition to that, there is an eating sound project that is quite similar and there is probably useful code like this one with wav2vec:

https://github.com/m3hrdadfi/soxan/blob/main/notebooks/Eating_Sound_Collection_using_Wav2Vec2.ipynb

On the other hand, non-ML could have even better results, as they follow simple rules that detect claps efficiently. For instance:

https://github.com/nikhiljohn10/pi-clap

https://github.com/xSparfuchs/clap-detection/blob/master/clap-detection.py

https://github.com/tom-s/clap-detector