In an Original Research paper published September 30, 2026 in Frontiers in Audiology and Otology, the authors separate training labels from scoring labels for learned auditory brainstem response (ABR) detectors.
The authors describe a secondary analysis of the earndb and Mendeley datasets, each containing recordings from eight adult listeners. They compare a label derived from correlation between repeated signal averages with published human annotations. They explicitly do not treat human judgment as physiological ground truth.
For learned detectors, the design crosses the training label with the scoring label. Holding training fixed while changing the scoring label asks how the same predictions are assessed. Holding the scoring label fixed while changing the training label asks what learning a different target does. Those comparisons answer different questions; a comparison that changes both choices cannot isolate either one.
The methods report leave-one-subject-out evaluation, withholding a listener from training when testing that listener. The authors also supply paired signal averages as separate inputs. Waveform sample counts therefore need to remain distinct from listener counts when describing this design.
The authors describe the analysis as exploratory, disclose decisions made after inspecting intermediate output and state that it was not preregistered. For audiology research readers, the useful starting questions are what target a detector learned and what target supplied its score. This methods account does not select a preferred detector or validate either label.
Original ENT Daily Dispatch methods account of Seyed-Ali Sadegh-Zadeh, Shayan Saadat, Mohammad Amin Barati, Septavera Sharvia and Delaram Jarchi, “Label construction can determine measured detector performance on small auditory brainstem response benchmarks,” Frontiers in Audiology and Otology, September 30, 2026, DOI 10.3389/fauot.2026.1952229, CC BY 4.0. Paraphrased; no endorsement implied.
An ABR detector comparison needs an explicit training target, scoring target and listener unit before its performance figures can be interpreted.
Go to the source.
Original research ↗CC BY 4.0 ↗This methods explainer describes public reference records. It does not establish study results, a product’s current authorization or treatment suitability.
