TruaceTracing the truth around AITuesday, July 21, 2026
Entertainment·P Space·Evidence-backed problem·Published 2026-07-20

LSTM and CNN-Based Detection of AI-Generated Classical Music From MIDI Features

Detecting AI-generated classical music is a growing challenge as artificial intelligence continues to improve its ability to compose pieces that closely resemble human compositions. This study explores the use of deep learning methods like LSTM and CNN to classify whether a set of classical music is generated by AI or humans. The classification is based on sequential features extracted from MIDI files using beat-based segmentation, capturing statistical data of pitch, velocity, and duration over segments. The mo…

TRV-2026-0447Peer-reviewedPermanent record — cite & verify
LSTM and CNN-Based Detection of AI-Generated Classical Music From MIDI Features

"Chopin Piano Concerto No. 2 In F Minor Op. 21 -midi Version- The Classical Orchestra The Classical Orchestra" by iClassicalCom is licensed under CC BY 2.0. To view a copy of this license, visit https://creativecommons.org/licenses/by/2.0/.

The quick read

A peer-reviewed study published November 10, 2025 tested LSTM and CNN models to classify classical music as AI-generated or human-composed using statistical pitch, velocity, and duration features extracted from MIDI files via beat-based segmentation. Trained on a dataset containing both types of compositions, the models were evaluated on primary and auxiliary test sets.

High classification accuracy matters for music attribution, copyright, and platform integrity as generative models improve, but the reported results are confined to the study's MIDI datasets and feature set. Generalization to broader repertoires, live audio, or newer generative systems remains untested and would require expanded data and additional musical features.

Main points
  • Study used beat-based segmentation to extract statistical pitch, velocity, and duration features from MIDI files.
  • Dataset included both AI-generated and human compositions with fine-tuning for optimal performance.
  • LSTM outperformed CNN, achieving 99.00% on primary test set and 98.70% on auxiliary dataset versus CNN's 97.00% and 97.10%.
  • Evaluation included confusion matrices and classification reports showing minimal misclassification rates.
Problem

AI systems can now compose classical pieces that closely resemble human compositions, making detection increasingly difficult.

The rundown

Researchers extracted sequential features from MIDI files using beat-based segmentation, capturing statistical data of pitch, velocity, and duration over segments, and trained LSTM and CNN classifiers on a mixed dataset of AI-generated and human compositions with fine-tuning.

The LSTM model achieved 99.00% accuracy on the primary test set and 98.70% on an auxiliary dataset, while the CNN achieved 97.00% and 97.10% respectively, with confusion matrices and classification reports showing minimal misclassification and indicating stronger performance for LSTM.

Sources

Reader signal

How should this claim be treated?

The debate