This projects consists in contributions to Deep Learning architectures for the task of voice assignment, applied to pitch transcription of audio files containing musical performances of a cappella quartets.
An application was deployed at HuggingFace. To use the app, click here. You only need to upload an audio file (any format) and the application will return MIDI, CSV and HDF5 containing approximated pitch transcription of four voices (Soprano, Alto, Tenor and Bass).
This repository contains the source codes for models, experiments, performance evaluation algorithms, graph plotting, among other things.