A Semantically Consistent Dataset for Data-Efficient Query-Based Universal Sound Separation

Query-based universal sound separation is fundamental to intelligent auditory systems, aiming to isolate specific sources from mixtures. Despite recent advances, existing methods continue to suffer from residual interference in complex acoustic scenes. This performance limitation stems largely from a data bottleneck: in-the-wild datasets contain weak labels and severe co-occurrence of events. These flaws induce models to learn spurious correlations between background noise and target categories instead of robust acoustic features. To address this, we propose an automated pipeline that eliminates co-occurrence of events by mining high-purity single-event segments from in-the-wild datasets via a semantically consistent synthesis protocol. Utilizing this pipeline, we constructed Hive, a high-quality synthetic dataset comprising 2.4k hours of raw audio. Experimental results demonstrate that, compared with the state-of-the-art model SAM-Audio which was trained on a huge dataset $\sim$500 times larger than Hive, certain open-source models trained on Hive achieve competitive separation accuracy and perceptual quality. Moreover, these models exhibited remarkable zero-shot generalization on out-of-distribution evaluation benchmarks. These findings highlight that prioritizing purity of supervised signals enables significant data efficiency, offering a new paradigm for training robust auditory foundation models with reduced computational costs. Code and dataset are available at https://cslikai.cn/Hive.

Imagine listening to a recording of a busy street and being able to remove everything except the dog barking, or keep only the rain. AI systems that pull individual sounds out of such mixtures are powerful, but they often leave traces of background noise behind. The reason hides in their training data. Most clips labeled "rain" also contain wind, traffic, or chatter, so the AI quietly learns those extras as part of what rain sounds like. The usual fix has been to throw vastly more data at the problem, training on millions of hours of audio with the computing budget to match. We took the opposite approach. Using modern AI models as automatic auditors, we sifted through twelve public audio collections and kept only short clips containing a single, clearly identifiable sound. We then recombined these clean clips into realistic mixtures, letting "typing" co-occur with "air conditioning" while rejecting impossible pairings such as "whales" with "city traffic." The resulting dataset, named Hive, is roughly 500 times smaller than the largest competitor, yet separation models trained on it match or beat that competitor and generalize well to unfamiliar recordings. Cleaner ingredients, it turns out, can replace bigger ovens.