HSBP-DB: A synchronised phonocardiography and arterial blood pressure dataset for multi-class hypertension staging under resting and exercise conditions.

HSBP-DB is a novel, annotated, and synchronised multimodal corpus of phonocardiography (PCG) and arterial blood pressure (BP) recordings, assembled to address the absence of a purpose-built, openly accessible benchmark for multi-class hypertension staging from cardiac acoustic signals. The dataset comprises 400 recordings acquired from 50 adult participants (32 male, 18 female; age: 22.70 ± 2.00 years; BMI: 23.82 ± 4.86 kg/m²) at the Gymnasium, Universiti Kuala Lumpur Malaysian Institute of Industrial Technology (UniKL MITEC), Johor, Malaysia. Each participant completed a repeated-measures protocol of five resting-baseline (RT) and three post-exercise (EX) sessions, yielding eight recordings per participant across two physiologically distinct haemodynamic conditions. PCG signals were acquired using the 3M Littmann CORE Digital Stethoscope (Model 8480), positioned at Erb's point (left sternal border, fourth intercostal space) and sampled at 4,000 Hz. Concurrent ground-truth BP measurements were obtained using the Contec ABPM50 oscillometric ambulatory monitor (ESH-IP 2010 validated; ±3 mmHg accuracy), applied to the left upper arm in accordance with AHA clinical measurement guidelines. A sequential reference-first protocol was enforced to ensure temporal alignment between the haemodynamic state encoded in the BP reference label and the corresponding PCG recording. Each recording is assigned a structured anonymisation identifier embedding the participant code, session index, systolic pressure (SBP), diastolic pressure (DBP), and mean arterial pressure (MAP). BP class labels were applied at the recording level according to the 2017 American Heart Association / American College of Cardiology (AHA/ACC) four-class schema - Normal, Elevated, Hypertension Stage 1 (HTN-1), and Hypertension Stage 2 (HTN-2) - yielding a class distribution of Normal: 170 (42.5%), Elevated: 78 (19.5%), HTN-1: 112 (28.0%), and HTN-2: 40 (10.0%). HSBP-DB is deposited exclusively as raw, uncompressed .wav PCG waveforms, organised into four class-stratified directories mirroring the AHA/ACC classification scheme. All recording-level metadata - participant identity, session index, and haemodynamic labels - are self-contained within each filename, eliminating dependency on external metadata files while preserving full label-to-signal traceability. HSBP-DB is publicly deposited in Mendeley Data (DOI: 10.17632/fwx556cgjy.1) and is made available to support reproducible research in cuffless BP estimation, PCG-based haemodynamic biomarker extraction, ordinal classification, and multimodal cardiovascular signal processing.