NEXTLab-ZJU/popular-hook

Popular Hooks

This is the dataset repository for the paper: Popular Hooks: A Multimodal Dataset of Musical Hooks for Music Understanding and Generation, in 2024 IEEE International Conference on Multimedia and Expo Workshops (ICMEW).

1. Introduction

Popular Hooks, a shared multimodal music dataset consisting of 38,694 popular musical hooks for music understanding and generation; this dataset has the following key features:

• Multimodal Music Data
• Accurate Time Alignment
• Rich Music Annotations

2. Modalities

• Midi
• Lyrics
• Video (Youtube link provided, you need to download it by yourself)
• Audio

3. High Level Music Information

• Melody
• Harmony
• Structure
• Genre
• Emotion(Russell's 4Q)
• Region

4. Dataset File Structure

• info_tables.xlsx: it contains a list describing the baisc information of each midi file (index, path, song name, singer, song url, genres, youtube url, youtube video start time and end time/duration, language, tonalities)
• midi/{index}/{singer_name}/{song_name}:
  • complete_text_emotion_result.csv: it contains the emotion class(4Q) which is predicted with the total lyrics of the song.
  • song_info.json: it contains the song's section info, theorytab DB url and genres info.
  • total_lyrics.txt: it contains the song's complete lyrics which is collected from music API(lyricsGenius, NetEase, QQMusic)
  • youtube_info.json: it contains the url of the song in Youtube, the start time and end time/duration of the video section.
  • ./{section}
    • {section}.mid: the section in midi format
    • {section}.txt: it contains the tonalites of the section.
    • {section}_audio_emotion_result.csv: it contains the emotion class(4Q) which is predicted with the audio of the section.
    • {section}_lyrics.csv: it contains the lyrics of the section.
    • {section}_midi_emotion_result.csv: it contains the emotion class(4Q) which is predicted with the midi of the section.
    • {section}_multimodal_emotion_result.csv: it contains the emotion class(4Q) which is selected from the multimodal emotions of the section.
    • {section}_text_emotion_result.csv: it contains the emotion class(4Q) which is predicted with the lyrics of the section.
    • {section}_video_emotion_result.csv: it contains the emotion class(4Q) which is predicted with the video of the section.

5. Demo