I am working on a project wherein I want to classify Tabla taalas(patterns) and I didn't find any dataset regarding it. I am recording them myself and I've ~500 data samples recorded. What model shall I use to classify the patterns if I have less than 500 samples and 6 classes?
Topic multiclass-classification deep-learning machine-learning
Category Data Science
500 samples for 6 classes is not so much. You should put aside about 100 samples for validation and 100 for testing, leaving 300 samples for training. I'm assuming that these drum loop are on the order of 1 second long (0.5-5 seconds). Then I would recommend trying a pretrained audio model, which typically have 1 second analysis windows. For example OpenL3, which is a powerful and easy to use audio embedding. It even has pretrained versions trained on audio. On top of the audio embeddings try a simple linear classifier like LogisticRegression or a RandomForest.
Geeks Mental is a community that publishes articles and tutorials about Web, Android, Data Science, new techniques and Linux security.