NumpyNetwork

Need a code to understand the theory behind Neural Networks? Come, I'll help you

Testing out - Decompose and Recompose input

Input - Sona Mutemix Lofi - Made by https://twitter.com/alaylays

Output over 100 epochs(100x100 image)

gif.mp4

Input - Eternal Return Black Survival - Chiara Voice Line

chiara_hacking_1_en.mp4

Output over 90 epochs

output1.mp4

output2.mp4

output3.mp4

output4.mp4

output5.mp4

output6.mp4

output7.mp4

output8.mp4

output9.mp4

output10.mp4

FUN FACT: Using a simple random gaussian noise with center 0 and std = 0.05 returned a similar output for both cases, but with audio in the left stereo only

References

Some classes from Didática Tech(PT-BR): https://didatica.tech/

Like this one

I must say that most tutorials are quite confusing

https://towardsdatascience.com/nothing-but-numpy-understanding-creating-binary-classification-neural-networks-with-e746423c8d5c

https://medium.com/@neuralthreads/binary-cross-entropy-loss-special-case-of-categorical-cross-entropy-loss-95c0c338d183

https://medium.com/@neuralthreads/categorical-cross-entropy-loss-the-most-important-loss-function-d3792151d05b

https://medium.com/analytics-vidhya/2d-convolution-using-python-numpy-43442ff5f381

https://towardsdatascience.com/types-of-convolutions-in-deep-learning-717013397f4d

https://automata88.medium.com/how-to-implement-the-softmax-derivative-independently-from-any-loss-function-ae6d44363a9d

https://en.wikipedia.org/wiki/Softmax_function

https://medium.com/arteos-ai/the-differences-between-sigmoid-and-softmax-activation-function-12adee8cf322

https://github.com/CaptainE/RNN-LSTM-in-numpy/blob/master/RNN_LSTM_from_scratch.ipynb

https://towardsdatascience.com/backpropagation-in-a-convolutional-layer-24c8d64d8509