This repo contains several models which map piano to orchestral scores as a way of simulating the work of a human orchestrator.
This project is inspired by the research done McGill University: https://hal.archives-ouvertes.fr/hal-01578292/document This study accumulated several datasets of piano and corresponding orchestral scores. Our models were trained on this data.
Our repo contains two types of sequence-to-sequence machine learning models. On the naveen2 branch, there is a model which uses a single RNN to read in the piano score data directly and generate an orchestral mapping, while the multipleRNN branch contains a model which is being trained on an instrument specific basis, with individual RNNs mapping each part in the orchestra.