PAC-KD: Pathology-Aware Compositional Knowledge Distillation for Respiratory Sound Classification

Li Xiao; Weiping Tu; Youqiang Zheng; Dan Chen

cs.SDOct 16, 2026

PAC-KD: Pathology-Aware Compositional Knowledge Distillation for Respiratory Sound Classification

Authors: Li Xiao, Weiping Tu, Youqiang Zheng, Dan Chen

Organizations: School of Computer Science, Wuhan University, Wuhan 430072, China

Abstract

Respiratory sound classification distinguishes four categories: Normal, Crackle, Wheeze, and Both (crackle and wheeze). Existing knowledge distillation transfers four-class predictions without explicitly exploiting shared respiratory events, potentially limiting the transfer of event-specific knowledge across individual abnormal classes and the Both class. We propose PAC-KD, a pathology-aware compositional knowledge distillation framework that learns crackle and wheeze presence through two dedicated specialists. Their complementary knowledge is transferred to a unified four-class student, enabling it to effectively capture pathology-specific attributes while preserving class-level discrimination. Furthermore, stable conflict calibration adaptively regulates specialist guidance according to its consistency with classification supervision, mitigating potential supervision conflicts. Experiments on ICBHI and SPRSound achieve Scores of 65.28% and 89.54%, respectively, outperforming strong knowledge distillation baselines on both datasets.

Figures & tables

Fig. 1: Overview of PAC-KD. (1) Two pathology specialists learn crackle and wheeze presence from four-class annotations. (2) Their frozen predictions supervise the corresponding student marginals through Bernoulli divergences $D_{C}$ and $D_{W}$ . (3) Stable conflict calibration smooths gradient compatibility and maps it to a bounded distillation weight $\lambda_{t}$ . Only the four-class student is retained at inference. Superscripts $\mathrm{T}$ and $\mathrm{S}$ denote teacher and student.

Method	Venue	ICBHI			SPRSound
Method	Venue	$S_{p}$ (%)	$S_{e}$ (%)	Score (%)	$S_{p}$ (%)	$S_{e}$ (%)	Score (%)
Conventional Learning
Patch-Mix CL [ 1 ]	Interspeech’23	81.66	43.07	$62.37^{\dagger}$	90.41 ±0.28	87.27 ±0.13	88.81 ±0.16
RepAugment [ 10 ]	EMBC’24	82.47	40.55	$61.51^{\dagger}$	92.52 ±1.42	84.58 ±1.47	88.55 ±0.56
BTS [ 11 ]	Interspeech’24	81.40	45.67	$63.54^{\dagger}$	92.42 ±1.70	85.24 ±1.33	88.83 ±0.62
Jeong et al. (PAFA) [ 7 ]	Interspeech’25	82.05	47.63	$\underline{64.84}^{\dagger}$	89.33 ±1.73	88.53 ±1.18	88.93 ±0.76

TABLE I: Comparison with existing methods on ICBHI and SPRSound. Results marked with

\dagger

are quoted from the corresponding publications; the marker applies to all three metrics of the corresponding dataset. For each metric, the best and second-best results are shown in bold and underlined text , respectively.

Label	ICBHI		SPRSound
Label	Train	Test	Train	Test
Normal	2,063	1,579	5,159	1,040
Crackle	1,215	385	961	83
Wheeze	501	649	506	305
Both	363	143	30	1
Total	4,142	2,756	6,656	1,429

TABLE II: Number of respiratory cycles in the ICBHI and harmonized SPRSound splits. SPRSound adopts an inter-patient test split.

Method	$S_{p}$ (%)	$S_{e}$ (%)	Score (%)
Student	82.05	47.63	$64.84^{\dagger}$
Student (Attribute Head)	$77.58_{\pm 2.21}$	$50.76_{\pm 1.86}$	$64.17_{\pm 0.26}$
Four-class KD	$78.71_{\pm 0.83}$	$49.48_{\pm 1.54}$	$64.10_{\pm 0.75}$
Four-class KD + SCC	$78.99_{\pm 1.64}$	$50.47_{\pm 1.42}$	$64.56_{\pm 0.67}$
Dual-specialist KD	$77.64_{\pm 3.37}$	$\textbf{52.39}_{\pm 2.49}$	$65.02_{\pm 0.44}$
PAC-KD	$79.35_{\pm 3.94}$	$51.21_{\pm 3.94}$	$\textbf{65.28}_{\pm 0.29}$

TABLE III: Ablation of teacher knowledge and stable conflict calibration (SCC) on ICBHI. All rows use the same student backbone and evaluation protocol. SCC indicates conflict-calibrated distillation.

Fig. 2: Class-wise attribute errors of a four-class teacher and two pathology-specific specialists in the shared crackle–wheeze attribute space. Lower is better .

Backbone	Method	$S_{p}$ (%)	$S_{e}$ (%)	Score (%)
CNN6	CE	83.05	31.88	$57.46^{\dagger}$
CNN6	PAC-KD	$76.68_{\pm 1.75}$	$\textbf{39.05}_{\pm 2.06}$	$\textbf{57.86}_{\pm 0.63}$
AST	CE	79.18	42.68	$60.98^{\dagger}$
AST	PAC-KD	$77.26_{\pm 4.68}$	$\textbf{45.43}_{\pm 3.17}$	$\textbf{61.35}_{\pm 0.89}$
PAFA	CE	82.05	47.63	$64.84^{\dagger}$
PAFA	PAC-KD	$79.35_{\pm 3.94}$	$\textbf{51.21}_{\pm 3.94}$	$\textbf{65.28}_{\pm 0.29}$

TABLE IV: Performance of PAC-KD with different student backbones on ICBHI.

Respiratory Sound Classification

Respiratory sound classification uses audio and machine learning to identify breathing events and support respiratory disease assessment.

See all 28 papers

Clinically Informed Acoustic-Aware Attribute Modeling for Respiratory Sound Classification
Oct 16, 2026Xinhong Li, Li Xiao, Ziyang Chen +2Respiratory Sound Classification Classification
School of Computer Science, Wuhan University, Wuhan, China · Hubei Key Laboratory of Multimedia and Network Communication Engineering, Wuhan University, Wuhan, China
APRIL: Ambiguity-Aware Expert Routing for Respiratory Sound Domain-Incremental Learning
Oct 20, 2026Xinhong Li, Weiping TuRespiratory Sound Classification Domain-Incremental Learning
School of Computer Science, Wuhan University, Wuhan, Hubei, China
RespFocus: Learning Temporal Evidence for Weakly Supervised Respiratory Sound Classification
Oct 11, 2026Yi Tian, Xiaohuan Chen, Yufei tang +5Respiratory Sound Classification Weakly Supervised Learning
Fuyao University of Science and Technology (fyust.edu.cn) · Xiamen University (xmu.edu.cn)

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

Clinically Informed Acoustic-Aware Attribute Modeling for Respiratory Sound Classification

Xinhong Li, Li Xiao, Ziyang Chen, Yuhong Yang+1

School of Computer Science, Wuhan University, Wuhan, China · Hubei Key Laboratory of Multimedia and Network Communication Engineering, Wuhan University, Wuhan, China

Respiratory Sound Classification Classification Multi-Label Classification

Open paper AskPDF arXiv

Automated respiratory sound classification (RSC) aims to identify normal and adventitious respiratory sounds, such as crackles and wheezes. Most existing methods formulate RSC as a flat four-class classification problem over norma, crackle, wheeze and both. However, unlike conventional classification tasks with semantically independent and mutually exclusive categories, the commonly used four-class RSC labels have a compositional structure and can be represented by two clinically meaningful sound attributes: crackle presence and wheeze presence. Existing flat classifiers do not explicitly exploit this structure, which may increase the learning burden by treating the four labels as independent targets. Meanwhile, crackles and wheezes exhibit distinct temporal characteristics, with sparse transient crackle events and sustained continuous wheeze patterns, making a single temporal aggregation strategy less suitable for modeling different adventitious sound evidence. To address these limitations, we propose an Acoustic-Aware Asymmetric Framework (AAAF) for RSC. AAAF reformulates four-class recognition as two sound-attribute detection tasks and recovers the final category through logical composition during inference. Built upon a pretrained BEATs encoder, AAAF employs branch-specific multi-layer fusion and asymmetric temporal aggregation to capture localized crackle evidence and sustained wheeze patterns. Auxiliary normal--abnormal supervision and a batch-adaptive dynamic loss are further introduced to improve global discrimination and robustness under imbalanced labels. Experiments on ICBHI and SPRSound demonstrate that AAAF improves overall RSC performance and generalizes across datasets, achieving a state-of-the-art four-class Score of 65.30% on ICBHI.

Abstract

Respiratory Sound Classification

Explore similar work

Clinically Informed Acoustic-Aware Attribute Modeling for Respiratory Sound Classification

APRIL: Ambiguity-Aware Expert Routing for Respiratory Sound Domain-Incremental Learning

RespFocus: Learning Temporal Evidence for Weakly Supervised Respiratory Sound Classification