AI Paper Atlas

Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning

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

Kai Li, Jintao Cheng, Chang Zeng, Zijun Yan, Helin Wang, Zixiong Su, Bo Zheng, Xiaolin Hu

ICML 2026 · Label-efficient and robust learning › Zero-/few-shot and meta-learning · approach: Zero-shot robustness and transfer · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add semantic-consistency constraint to mixture construction (ablation)AudioSep · Hive test set (175k mixtures)SDR (dB): 3.12 -> 4.12Improved
add semantic-consistency constraint to mixture construction (ablation)AudioSep · Hive test set (175k mixtures)OQ: 2.96 -> 3.11Improved
add semantic-consistency constraint to mixture construction (ablation)FlowSep · Hive test set (175k mixtures)OQ: 2.64 -> 2.79Improved
add semantic-consistency constraint to mixture construction (ablation)FlowSep · Hive test set (175k mixtures)CLAP-T: 0.13 -> 0.17Improved
train AudioSep on Hive instead of original corpusAudioSep · Hive test set (zero-shot, avg 2-5 mix)SDR (dB): 2.37 -> 5.67Improved
train FlowSep on Hive instead of original corpusFlowSep · Hive test set (zero-shot, avg 2-5 mix)OQ: 2.76 -> 3.05Improved
train AudioSep on Hive (2.4k h) instead of original (14.1k h)AudioSep · USS-BenchSDR (dB): -1.86 -> 2.29Improved
train AudioSep on Hive (2.4k h) instead of original (14.1k h)AudioSep · MUSDB18-HQSDR (dB): -1.01 -> 1.36Improved
train AudioSep on Hive instead of original corpusAudioSep · VGGClean_evalSDR (dB): 8.67 -> 7.61Hurt
train AudioSep on Hive instead of original corpusAudioSep · VGGClean_evalOQ: 3.42 -> 3.44Improved
train on Hive to reduce co-occurrence shortcut gapAudioSep · controlled paired co-occurring vs decorrelated mixturesSDR gap (dB): -1.41 -> -0.39Improved
train FlowSep on Hive to reduce shortcut gapFlowSep · controlled paired co-occurring vs decorrelated mixturesOQ gap: -0.23 -> -0.10Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy