This repository contains the official implementation of our ICASSP 2025 paper

"Towards an Integrated Approach for Expressive Piano Performance Synthesis from Music Scores"

by Jingjing Tang, Erica Cooper, Xin Wang, Junichi Yamagishi, and György Fazekas.

• m2a/
  Contains the implementation for fine-tuning the MIDI-to-audio synthesis model using the ATEPP dataset, as well as the baseline model setup.

• m2m/
  Includes the expressive performance rendering model, designed to generate expressive performance MIDI files from symbolic music scores.

• objective_eval/ Scripts to run objective evaluation of synthesised MIDI Performances

Please refer to the README.md files inside each subdirectory (m2a/ and m2m/) for detailed instructions on inference and generation of target MIDI or audio outputs.

As for reproducting objective evaluation results of the m2a model, please refer to the objective_eval/README.md file. For the m2m model, the matrix could be reproduced by running the evaluation script in the m2m/ directory.

We also provide a colab notebook for quick testing of both models: Colab Notebook.

The dataset for training the m2m model and all the checkpoints could be downloaded from Zenodo. For the dataset used to finetune the m2a model, please contact the authors directly.

You can listen to the demo samples on our project page.

Jingjing Tang: [email protected]

The code is licensed under Apache License Version 2.0, following ESPnet. The pretrained model is licensed under the Creative Commons License: Attribution 4.0 International http://creativecommons.org/licenses/by/4.0/legalcode

This work is supported by both the UKRI Centre for Doctoral Training in Artificial Intelligence and Music (grant number EP/S022694/1), and the National Institute of Informatics in Japan. J.Tang is a research student supported jointly by the China Scholarship Council [grant number 202008440382] and Queen Mary University of London. E. Cooper conducted this work while at NII, Japan and is currently employed by NICT, Japan.

@INPROCEEDINGS{10890623,
  author={Tang, Jingjing and Cooper, Erica and Wang, Xin and Yamagishi, Junichi and Fazekas, György},
  booktitle={ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)}, 
  title={Towards An Integrated Approach for Expressive Piano Performance Synthesis from Music Scores}, 
  year={2025},
  pages={1-5},
  doi={10.1109/ICASSP49660.2025.10890623}}