Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Aria-MIDI: A Dataset of Piano MIDI Files for Symbolic Music Modeling

Created by
  • Haebom

Author

Louis Bradshaw, Simon Colton

Outline

This paper introduces "ARIA-MIDI", a large-scale dataset of piano performance audio collected from the Internet and converted to MIDI files. Approximately 100,000 hours of audio were converted to over 1 million MIDI files through a multi-stage pipeline that automatically collects and evaluates audio sources using language models, and then uses audio classifiers to remove and segment unnecessary parts. Statistical analysis and metadata tag information of the dataset are also provided, and the dataset is open to the public on Github.

Takeaways, Limitations

•
Takeaways:
◦
Contributing to research on music information retrieval, generation, and analysis by providing a large-scale piano performance MIDI dataset
◦
Validating the effectiveness of automated data collection and preprocessing pipelines
◦
Increased usability of datasets by providing various metadata
•
Limitations:
◦
Lack of detailed evaluation of the quality and accuracy of the sound source.
◦
Lack of analysis of bias in the dataset (e.g. over-representation of certain genres or performers)
◦
Lack of specific descriptions of errors that may occur during the automatic transcription process.
👍
Made with Slashpage