AlayaLab/FlowSep-hive

1---2license: apache-2.03language:4- en5tags:6- audio7- sound-separation8- audio-to-audio9- flowsep10datasets:11- ShandaAI/Hive12---13 14# FlowSep-hive15 16## Model Description17 18**FlowSep-hive** is a data-efficient, query-based universal sound separation model trained on the [Hive dataset](https://huggingface.co/datasets/ShandaAI/Hive). By leveraging the high-quality, semantically consistent Hive dataset, this model achieves competitive separation accuracy and perceptual quality comparable to state-of-the-art models (such as SAM-Audio) while utilizing only a fraction (~0.2%) of the training data volume.19 20This model is developed by **Shanda AI Research Tokyo** and is introduced in the paper: [A Semantically Consistent Dataset for Data-Efficient Query-Based Universal Sound Separation](https://arxiv.org/abs/2601.22599).21 22## Model Details23 24- **Model Type:​** Query-Based Universal Sound Separation25- **Language(s):​** English (for text queries)26- **License:​** Apache 2.0 (Please update if different)27- **Trained on:​** [ShandaAI/Hive](https://huggingface.co/datasets/ShandaAI/Hive) (2,442 hours of raw audio, 19.6M mixtures)28- **Paper:​** [arXiv:2601.22599](https://arxiv.org/abs/2601.22599)29- **Code Repository:​** [GitHub - ShandaAI/Hive](https://github.com/ShandaAI/Hive)30 31## Uses32 33The model is intended for universal sound separation tasks, allowing users to extract specific sounds from complex audio mixtures using multimodal prompts (e.g., text descriptions or audio queries).