NEXTLab-ZJU/popular-hook

---
tags:
- music
- midi
- emotion
size_categories:
- 10K<n<100K
---
# 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
<img src='https://huggingface.co/datasets/NEXTLab-ZJU/popular-hook/resolve/main/imgs/popular_hooks_demo.png'>