Dataset · AlayaLab
Dataset for text-prompted sound separation
AlayaLab/Hive
A multimodal corpus of millions of examples for training universal query-based audio separation systems.
What's inside
The dataset contains 1 to 10 million records combining audio, text queries, and tabular data in English. Created by researchers from Tsinghua University, Shanda AI, and Johns Hopkins University for the audio-to-audio task. The materials are accompanied by a 2026 research paper, code, and a demo.
What you can do with it
- Training Universal Sound Separation models based on text descriptions
- Testing multimodal audio systems for semantic consistency
- Tasks
- Audio → audio
- Languages
- English
- Size
- 1M–10M rows
- Downloads this month
- 271
- Likes
- 53
How to load it
from datasets import load_dataset
ds = load_dataset("AlayaLab/Hive")