可学习多分辨率 ALE 睡眠声音事件检测
将 ALE 逐帧置信图直接注入 Transformer 注意力与特征调制。
为什么重要
为麦克风筛查睡眠呼吸障碍提供了可复现的新型 SED 架构思路。
阅读原文arxiv.org
AI 解读(根据原文生成)
论文提出 ALE-Confidence Fusion Sound Event Detection(ACF-SED),用于仅依赖麦克风的睡眠呼吸相关声音事件检测。方法不把 Adaptive Line Enhancer(ALE)仅作为预处理,而是将其逐帧置信图注入 Transformer 注意力偏置和特征调制;同时用 Multi-Resolution ALE Bank 聚合三个不同去相关延迟的并行 NLMS 滤波器,并用 Confidence-Guided Cross-Path block 通过门控交叉注意力耦合事件与噪声流。论文称其在 Audio-Polygraphy Dataset for Sleep Apnea Analysis 上达到 Event-F1、Segment-F1 和 PSDS 的 state-of-the-art,并支持端到端 AHI 估计。论文已在 arXiv 发布,许可为 CC BY 4.0。
要点(AI 据原文整理,数字均已与原文核对)
- ACF-SED injects the per-frame confidence map of an Adaptive Line Enhancer into Transformer attention biasing and feature modulation.
- A Multi-Resolution ALE Bank aggregates three parallel NLMS filters at distinct decorrelation delays.
- A Confidence-Guided Cross-Path block couples event and noise streams via gated cross-attention.
- ACF-SED achieves state-of-the-art detection across Event-F1, Segment-F1, and PSDS.
- Authors: Chanwoo Park, Chanwoo Kim.