This repo is the code of an ETH project from Luca Naterop and Sandro Giacomuzzi which aims to synthesize Audio with LSTMs.

/tf-rnn: The main code of this project

/csv-rnn: Torch LSTM implementation in

/char-rnn: The code for a character language model in torch which was trained with Darwin and the Koran.

/char-rnn-results: contains sample text of 100'000 characters generated from those networks as well as the corresponding checkpoints that allows you to generate more text.

/Latest_AudioFeature: MATLAB scripts to generate spectrogram csv files from Audio and vice versa

Paper (Work in Progress): https://www.overleaf.com/read/cnkdtvbtfsyd

Research Plan: https://www.overleaf.com/3808219jvshrb

Character language models: https://github.com/karpathy/char-rnn