This project uses GPT2 in order to automatically transcribe classical guitar audio files into sheet music. This is done by first training a model to output a vector representing a piano roll given a mel spectrogram. The MAESTRO dataset found here is made from pairs of audio and MIDI data (aligned with ~3ms accuracy) from multiple years of piano recitals, totaling 200 hours. This was converted into input and output pair tensors with dataset_creation.py
Alphatab will be used for tab visualization.
