Spoken Language Recognition

This notebook trains a convolutional neural network to classify audio files of voice recordings into the languages that were spoken. The dataset I used contained 66.000 files across 176 languages. I found it on TopCoder (https://goo.gl/G5XBJl). I liked the idea behind this problem, because it's very hard for humans to do. It's intersting to see that CNNs perform well on problems where intuition doesn't get you anywhere.

I included a saved version of my pretrained model, which evaluates to an accuracy of 98,79%. Further notes on development can be seen in the Jupyter Notebook.