POP909: A Pop-song Dataset for Music Arrangement Generation
POP909 dataset offers 909 pop songs' piano arrangements, aiding music arrangement generation research.
Key Findings
Methodology
The POP909 dataset comprises multiple versions of piano arrangements for 909 pop songs, with detailed annotations of melody, tempo, and chords. Professional musicians ensure high quality, and MIR algorithms assist in annotations.
Key Results
- The POP909 dataset excels in piano arrangement generation tasks, providing rich melody and accompaniment data to support deep learning model training and evaluation.
- Experiments show significant performance improvements in melody and accompaniment generation tasks using the POP909 dataset.
- The dataset's multi-version feature offers broader application scenarios for researchers.
Significance
The introduction of the POP909 dataset provides a crucial foundational resource for the music arrangement generation field. It fills the gap in existing datasets for arrangement tasks, supporting more precise model evaluations and practical generation results.
Technical Contribution
POP909 surpasses existing datasets in data quality and annotation precision, offering precise time alignment and detailed musical information annotations to support complex music generation tasks.
Novelty
POP909 is the first high-quality dataset focused on pop song piano arrangements, providing multiple arrangement versions and detailed annotations, filling a gap in the music generation field.
Limitations
- The dataset is primarily focused on pop music, which may not be applicable to other music genres.
- The annotated chord and tempo information might have some inaccuracies.
Future Work
Future research can explore the dataset's application in other music genres and develop more complex arrangement generation models.
AI Executive Summary
Music arrangement generation is a vital subfield of automatic music generation, involving the reconstruction and re-conceptualization of original melodies and chords. Existing arrangement models, although promising, lack refined datasets for better evaluation and practical application.
The introduction of the POP909 dataset addresses this gap. It includes multiple versions of piano arrangements for 909 pop songs, created by professional musicians to ensure high quality. The dataset provides detailed annotations of melody, tempo, and chords, with some annotations assisted by MIR algorithms, offering a rich resource for researchers.
Experiments demonstrate significant performance improvements in melody and accompaniment generation tasks using the POP909 dataset. Its multi-version feature offers broader application scenarios, contributing to the advancement of the music arrangement generation field.
Deep Analysis
Background
Music arrangement generation is a crucial task in automatic music generation, involving the reconstruction of original melodies and chords. Despite some progress with existing generative models, there is a lack of refined datasets for evaluation and practical application.
Core Problem
Existing datasets lack precise time alignment and detailed musical information annotations for arrangement tasks, limiting model evaluation and the practical application of generation results.
Innovation
The POP909 dataset addresses the shortcomings of existing datasets in arrangement tasks by providing multiple arrangement versions and detailed annotations, supporting more precise model evaluations and practical generation results.
Methodology
- �� Data Collection: Professional musicians create piano arrangements
- �� Data Annotation: Manual tempo curve labeling, MIR algorithm-assisted chord annotation
- �� Data Format: MIDI format, including melody and accompaniment information
Experiments
Experiments use the POP909 dataset for piano arrangement generation tasks, evaluating model performance in melody and accompaniment generation. Results indicate the dataset supports more precise model evaluation.
Results
Experiments show significant performance improvements in melody and accompaniment generation tasks using the POP909 dataset, supporting more complex music generation tasks.
Applications
The POP909 dataset can be used for music arrangement generation, melody accompaniment generation, and supports deep learning model training and evaluation.
Limitations & Outlook
The dataset is primarily focused on pop music, which may not be applicable to other music genres. Annotated chord and tempo information might have some inaccuracies.
Plain Language Accessible to non-experts
Imagine you're in a kitchen cooking. The POP909 dataset is like a detailed cookbook offering various recipes for different dishes. Each dish has detailed steps and ingredient lists, similar to the melody, tempo, and chord annotations in the dataset. With these recipes, you can try different methods to create your own delicious dishes.
ELI14 Explained like you're 14
Imagine playing a music game; the POP909 dataset is like the music library in the game, filled with piano versions of popular songs. You can choose different songs and try different arrangement styles, just like unlocking new levels in a game—fun and exciting!
Glossary
MIDI (Musical Instrument Digital Interface)
MIDI is a digital music protocol for transmitting musical information between instruments and computers.
In POP909, all arrangements are stored in MIDI format.
MIR (Music Information Retrieval)
MIR is a technique for extracting musical features from audio, used for analysis and annotation.
Used to annotate tempo and chord information in the POP909 dataset.
Arrangement
Arrangement is the process of re-conceptualizing and organizing existing musical works.
The POP909 dataset focuses on piano arrangements of pop songs.
Melody
Melody is the most recognizable part of music, composed of a series of notes.
The POP909 dataset provides melody information for each song.
Chord
A chord is a musical element composed of multiple notes sounding simultaneously, often used for accompaniment.
The POP909 dataset provides detailed chord annotations.
Open Questions Unanswered questions from this research
- 1 How can the POP909 dataset be applied to arrangement generation in other music genres?
- 2 What improvements can be made in MIR algorithms for annotation accuracy?
Applications
Immediate Applications
Music Arrangement Generation
Researchers can use the POP909 dataset for research and model evaluation in arrangement generation tasks.
Long-term Vision
Automated Music Creation
The POP909 dataset can serve as a foundation for automated music creation systems, advancing the automation of music composition.
Abstract
Music arrangement generation is a subtask of automatic music generation, which involves reconstructing and re-conceptualizing a piece with new compositional techniques. Such a generation process inevitably requires reference from the original melody, chord progression, or other structural information. Despite some promising models for arrangement, they lack more refined data to achieve better evaluations and more practical results. In this paper, we propose POP909, a dataset which contains multiple versions of the piano arrangements of 909 popular songs created by professional musicians. The main body of the dataset contains the vocal melody, the lead instrument melody, and the piano accompaniment for each song in MIDI format, which are aligned to the original audio files. Furthermore, we provide the annotations of tempo, beat, key, and chords, where the tempo curves are hand-labeled and others are done by MIR algorithms. Finally, we conduct several baseline experiments with this dataset using standard deep music generation algorithms.