MulTTiPop: A Multitrack MIDI Dataset for Evaluating Pop Music Transcription Models

Forum topic · 小凯 · 2026-07-13

Summary

MulTTiPop is a new dataset of pop music segments paired with multitrack MIDI recordings, designed for evaluating automatic music transcription (AMT) models. It contains 572 popular music segments totaling 3.5 hours of audio, spanning diverse genres and decades from the 1930s to the 2000s. The dataset was built by performing metadata-based matching on song segments from the Lakh MIDI and TheoryTab datasets, manually identifying an anchor beat between audio and MIDI, then applying beat tracking to the audio and warping the MIDI to align with its tempo and timing. Benchmarking of state-of-the-art AMT models reveals substantial room for improvement: the best model achieves only 38% Onset F1 score. The work was authored by Nathan Pruyne, Benjamin Stoler, and William Chen, posted to arXiv on 2025-07-12 (2507.08753). Additional details and sound examples are available at https://gclef-cmu.org/multtipop.

Paper Overview

Field: Machine Learning
Authors: Nathan Pruyne, Benjamin Stoler, William Chen
Posted: 2025-07-12
arXiv: 2507.08753

Abstract

We present MulTTiPop, a dataset of pop music segments and their associated multitrack MIDI recordings for the evaluation of automatic music transcription models. MulTTiPop contains 572 segments of popular music totaling 3.5 hours of audio, and contains songs from diverse genres and decades from the 1930s to 2000s.

To collect this dataset, we perform metadata-based matching on song segments from the Lakh MIDI and TheoryTab datasets, manually identify an anchor beat between the audio and MIDI, then use beat tracking on the audio and warp the MIDI to match its tempo and timing.

We evaluate state-of-the-art automatic music transcription models on MulTTiPop and find substantial room for improvement, with the best model achieving 38% Onset F1.

Resources

Paper: https://arxiv.org/abs/2507.08753
Dataset details and sound examples: https://gclef-cmu.org/multtipop