RespFocus: Learning Temporal Evidence for Weakly Supervised Respiratory Sound Classification

Yi Tian; Xiaohuan Chen; Yufei tang; Xinyu Lin; Hongliang Li; Chengbin Hou; Degen Huang; Rongshan Yu

cs.LGOct 11, 2026

RespFocus: Learning Temporal Evidence for Weakly Supervised Respiratory Sound Classification

Authors: Yi Tian, Xiaohuan Chen, Yufei tang, Xinyu Lin, Hongliang Li, Chengbin Hou, Degen Huang, Rongshan Yu

Organizations: Fuyao University of Science and Technology (fyust.edu.cn) · Xiamen University (xmu.edu.cn)

Abstract

Respiratory auscultation provides a low-cost means of assessing pulmonary function, yet transient crackles and sustained wheezes can be difficult to recognize in noisy clinical recordings. Conventional deep learning classifiers often operate as black boxes, offering limited insight into the temporal acoustic evidence supporting their predictions. We present RespFocus, a weakly supervised framework that uses multiple instance learning (MIL) to learn inspectable temporal evidence from cycle-level labels and integrate it into classification. We also introduce PedResp2026, a multicenter pediatric respiratory sound cohort collected in routine clinical environments with adjudicated cycle-level annotations. In classification experiments, RespFocus achieves four-class Scores of 69.55% on public ICBHI 2017 and 72.91% on PedResp2026. Evidence diagnostics and matched temporal interventions support the functional contribution of the learned evidence to model predictions. Together, these findings highlight RespFocus's potential as a tool for automated respiratory sound diagnosis.

Figures & tables

Figure 1: Illustrative comparison of conventional cycle classification (left) and RespFocus (right) with cycle intervals and temporal evidence. Amber/teal denote crackle/wheeze evidence. N/C/W/B denote Normal/Crackles/Wheezes/Both.

Figure 2: Illustrative RespFocus overview. (a) Dual-stream fusion and guidance from shared abnormal evidence. (b) Subtype MIL pooling and compositional decoding. (c) Shared abnormal, crackle, and wheeze evidence. (d) Optional text summaries with temporal references.

Method	Backbone	Venue	ICBHI 2017			PedResp2026
			Sc	Se	Sp	Sc	Se	Sp
CE ( Moummad and Farrugia, 2023 )	CNN6	WASPAA’23	55.19	40.32	70.07	50.00	20.00	80.00
SCL ( Moummad and Farrugia, 2023 )	CNN6	WASPAA’23	57.57	39.21	75.93	61.92	46.97	76.86
AFT on Mixed ( Kim et al., 2023 )	AST	NeurIPS W’23	61.80	42.86	80.73	65.31	47.88	82.75
Fine-tuning ( Bae et al., 2023 )	AST	INTERSPEECH’23	59.52	41.90	77.13	57.08	27.88	86.27
Patch-Mix CL ( Bae et al., 2023 )	AST	INTERSPEECH’23	62.34	43.02	81.67	68.41	52.12	84.71

Table 1: Four-class classification (%). Sc denotes Score. RespFocus reports five-run means. Bold/underlined values mark best/second-best results.

Figure 3: Five-run results by metric (rows) and method (columns) for ICBHI 2017 (solid black lines, filled circles) and PedResp2026 (colored dashed lines, open diamonds). Right summaries show mean

\pm

sample SD. Run indices do not imply paired initialization across methods.

Variant	TSCA	Evidence	Binary pair	Asym. pooling	Sp (%)	Se (%)	Score (%)
Flat AST classifier	$\times$	$\times$	$\times$	$\times$	$76.40\pm 6.35$	$43.17\pm 4.07$	$59.79\pm 4.47$
Dual-stream global pooling	$\checkmark$	$\times$	$\times$	$\times$	$79.40\pm 5.67$	$45.40\pm 3.48$	$62.40\pm 4.11$
Evidence pathway + flat four-way	$\checkmark$	$\checkmark$	$\times$	$\times$	$81.07\pm 5.91$	$48.25\pm 3.52$	$64.66\pm 4.28$
Mean pooling	$\checkmark$	$\checkmark$	$\checkmark$	$\times$	$82.67\pm 5.48$	$50.95\pm 3.14$	$66.81\pm 2.85$
Full RespFocus	$\checkmark$	$\checkmark$	$\checkmark$	$\checkmark$	$83.07\pm 8.50$	$56.03\pm 4.07$	$69.55\pm 4.92$

Table 2: Progressive ICBHI 2017 four-class ablation (%, five-run mean

\pm

SD).

Figure 4: Illustrative waveform interventions with within-cycle min–max-normalized evidence (left). Right panels show deletion $D_{q}(b)=p_{q}(x)-p_{q}(x_{\mathrm{delete},b})$ and retention $R_{q}(b)=p_{q}(x_{\mathrm{retain},b})$ for subtype $q\in\{C,W\}$ and cycle-duration fraction $b$ .

Figure 5: Two-cycle illustration of a wheeze-dominant case.

Explore similar work

Resp-Agent: An Agent-Based System for Multimodal Respiratory Sound Generation and Disease Diagnosis
Feb 16, 2026Pengfei Zhang, Tianxin Xie, Minghao Yang +1Respiratory Sound Classification Audio Generation
The Hong Kong University of Science and Technology (Guangzhou)
Lung-SRAD: Spectral-Aware Regularized Audio DASS with Dual-Axis Patch-Mix Contrastive Learning for Respiratory Sound Classification
Jun 10, 2026Hemansh Shridhar, Miika Toikkanen, June-Woo KimContrastive Learning Spectral Regularization
RSC LAB, MODULABS, Republic of Korea · Department of Electronic Engineering, Wonkwang University, Republic of Korea · AI Convergence Research Institute, Wonkwang University, Republic of Korea
Adaptive Test-Time Scaling for Zero-Shot Respiratory Audio Classification
Apr 14, 2026Tsai-Ning Wang, Herman Teun den Dekker, Lin-Lin Chen +2Respiratory Sound Classification Test-Time Scaling
Eindhoven Artificial Intelligence Systems Institute, The Netherlands

Feb 16, 2026J/K move · Enter open · S save

Resp-Agent: An Agent-Based System for Multimodal Respiratory Sound Generation and Disease Diagnosis

Pengfei Zhang, Tianxin Xie, Minghao Yang, Li Liu

The Hong Kong University of Science and Technology (Guangzhou)

Respiratory Sound Classification Audio Generation Synthetic Data Augmentation Multimodal Classification

Open paper AskPDF arXiv

Deep learning-based respiratory auscultation is currently hindered by two fundamental challenges: (i) inherent information loss, as converting signals into spectrograms discards transient acoustic events and clinical context; (ii) limited data availability, exacerbated by severe class imbalance. To bridge these gaps, we present Resp-Agent, an autonomous multimodal system orchestrated by a novel Active Adversarial Curriculum Agent (Thinker-A

^2

CA). Unlike static pipelines, Thinker-A

^2

CA serves as a central controller that actively identifies diagnostic weaknesses and schedules targeted synthesis in a closed loop. To address the representation gap, we introduce a modality-weaving Diagnoser that weaves clinical text with audio tokens via strategic global attention and sparse audio anchors, capturing both long-range clinical context and millisecond-level transients. To address the data gap, we design a flow matching Generator that adapts a text-only Large Language Model (LLM) via modality injection, decoupling pathological content from acoustic style to synthesize hard-to-diagnose samples. As a foundation for this work, we introduce Resp-229k, a benchmark corpus of 229k recordings paired with LLM-distilled clinical narratives. Extensive experiments demonstrate that Resp-Agent consistently outperforms prior approaches across diverse evaluation settings, improving diagnostic robustness under data scarcity and long-tailed class imbalance. Our code and data are available at https://github.com/zpforlove/Resp-Agent.

Abstract

Explore similar work

Resp-Agent: An Agent-Based System for Multimodal Respiratory Sound Generation and Disease Diagnosis

Lung-SRAD: Spectral-Aware Regularized Audio DASS with Dual-Axis Patch-Mix Contrastive Learning for Respiratory Sound Classification

Adaptive Test-Time Scaling for Zero-Shot Respiratory Audio Classification