Standard Operating Procedure for Labeling

The below is Hyfe’s Standard Operating Procedure (SOP) for understanding, identifying, and labeling cough and cough-like acoustic events within continuous audio recordings. It forms part of Hyfe’s formal Validation Documentation (2023) and establishes the methodological foundation for producing high-quality, standardized ground-truth datasets used in algorithm development and performance verification.

This SOP is a comprehensive labeling manual that provides:

1. A Scientific Framework for Cough Physiology and Acoustics: The document begins with detailed explanations of cough anatomy and physiology—covering the pharynx, larynx, epiglottis, and vocal cords—and describes the acoustic waveforms associated with expulsive and vocal phases of coughs (pages 1–8). These diagrams and examples ensure that labelers understand not only what a cough sounds like, but why it sounds as it does biologically.
2. A Rigorous Multi-Tier Labeling System: The SOP defines explicit labeling categories (COUGH, THROAT_CLEAR, SNEEZE, CRY) and a structured tag system (Far, Not Sure) to capture acoustic ambiguity or distance effects (pages 9–11). These categories and tags are consistently illustrated using examples and spectrograms.
3. A Detailed Procedure for Consistent Annotation Using the Hyfe Continuous Labeling Web App: The document includes step-by-step instructions for opening, navigating, and labeling audio within Hyfe’s dedicated web application (pages 21–27). It prescribes zoom levels, selection methods, waveform inspection strategies, and use of spectrograms to ensure precise marking of beginnings and endings of events.
4. Extensive Examples and Edge-Case Demonstrations: Many pages contain annotated cough, throat-clear, and sneeze examples (pages 11–20), reflecting real-world acoustic complexity. These examples serve as training material and calibration references for human annotators.
5. A Formal Assessment and Competency Check for Labelers: The SOP includes a self-assessment questionnaire (pages 28–29) and defines a pass threshold (95%) for final evaluation before labelers are certified to contribute to gold-standard datasets.

Why This Document Is Critical for Validation

1. It Ensures Ground Truth Is Consistent, Reproducible, and Scientifically Defensible: All accuracy assessments of Hyfe’s cough detection algorithms depend on a high-quality ground truth—i.e., precise timestamps and classifications of cough events. This SOP provides the instructions that guarantee consistency in: (i) identifying expulsive vs. vocal cough phases, (ii) distinguishing coughs from throat clears, sneezes, or mechanical sounds, (iii) determining episode boundaries and zoom-level expectations, and (iv) applying ambiguity and distance tags uniformly. Because algorithm performance metrics (sensitivity, specificity, PPV, etc.) are only as reliable as the annotations they are compared against, the SOP ensures that human labeling itself is validated, controlled, and standardized.
2. It Defines the Gold-Standard Methodology Used Across All Validation Studies: As stated explicitly in the SOP (page 21), these annotations serve as the “gold standard by which Hyfe’s algorithms are compared.” This means: (i) Every validation dataset is created using the same procedures. (ii) Inter-annotator variability is minimized. (iii) Regulatory-grade reproducibility is ensured. Without this SOP, algorithm performance would risk being evaluated against inconsistent or subjective interpretations of cough sounds.
3. It Provides Traceability and Auditability Required in Pharma-Facing Validation: Because this SOP is structured, version-controlled, and includes a competency review, it supports: (i) traceable labeling workflows, (ii) auditor-verifiable labeling practices, (iii) documented human-factors controls, and (iv) clear justification for the reliability of the ground truth used in accuracy studies. These characteristics are essential for medical-device or pharma partners who require assurance of methodological rigor.
4. It Reduces Systematic Error and Bias in Annotation: By defining what a cough is, how it appears acoustically, and exactly how labelers should mark it, the SOP reduces: (i) false positives from similar explosive sounds, (ii) false negatives from subtle or distant coughs, (iii) bias resulting from inconsistent zoom levels or labeling strategies. This improves the quality of training and validation datasets, enabling more accurate and generalizable algorithmic performance.

This document is a core validation asset for Hyfe’s cough detection system. It is not merely a labeling guide—it is a scientifically grounded, operationally rigorous methodology for generating reliable, reproducible ground truth, upon which all accuracy assessments, performance claims, and regulatory-grade validations ultimately depend.

Labeling SOP