During some research on machine learning and neural networks, I came across one really cool video by SomethingUnreal. In this clip of roughly 14 minutes, you can see how a neural network attempts to generate English speech based on raw waveform training data. A lot of gibberish happens along the way but the results are highly fascinating. What do you think?

YouTube: Neural Network Tries to Generate English Speech (RNN/LSTM)

Photo credit: Surian Soosay