drscotthawley / panotti

A multi-channel neural network audio classifier using Keras
MIT License
271 stars 72 forks source link

GTZAN dataset #37

Open sarbjit-longia opened 6 years ago

sarbjit-longia commented 6 years ago

I am trying to train on GTZAN dataset. The accuracy is around 50% which seems very low. I have a question, the mel spectrogram has dimension of 96x2584 for each audio sample. Shall i use the whole sample as one "image" for the CNN network or do I need to divide the audio file into samples like 2048 and use CNN on that one.