Accelerating the search for antimicrobial peptides with Google Antigravity
deepmind.google·2h ago
AI Research
arXiv CS.AI·1d ago·5 min read
Gugak-VocalSet: A Multi-Genre Studio-Quality Dataset of Korean Traditional Singing with Sigimsae Annotations
TL;DR
Gugak-VocalSet is the first public dataset of Korean traditional singing, featuring 24.5 hours of audio with detailed sigimsae annotations. It enables advanced research in vocal expression analysis and classification.
✦ Why It Matters
Explore Gugak-VocalSet to enhance your projects in music analysis and AI-driven vocal research.
Key Takeaways
- Utilize Gugak-VocalSet for training models in vocal expression analysis.
- Leverage sigimsae annotations for improved pitch contour recognition.
- Explore genre and mood tags for diverse music classification tasks.
- Conduct experiments comparing domain-informed models with general music models.
Full Summary
Gugak-VocalSet introduces a comprehensive audio dataset focused on Korean traditional singing, comprising 24.5 hours of solo vocal recordings across 305 tracks. These recordings, performed by 20 professional singers, are meticulously annotated with sigimsae, a structured system for vocal pitch contours.
The dataset includes genre, lyrics, and mood tags, along with a unified sigimsae ontology developed by 14 expert annotators. It features approximately 68,000 audio segments with detailed annotations, allowing for nuanced analysis.
Baseline experiments demonstrate that specialized pitch representations outperform general music models in sigimsae classification tasks. The findings suggest that integrating domain-specific features can enhance event detection in traditional music.