Extended High Frequency Hearing May Signal Speech-in-Noise Problems Before Standard Tests Do

Difficulty understanding speech in noisy rooms is one of the most common complaints that brings people into audiology clinics, and it often persists even in patients whose standard hearing test looks perfectly normal. A new retrospective study of 605 adults, published in Frontiers in Neuroscience, reports that hearing sensitivity at extended high frequencies above 8 kilohertz is a stronger predictor of speech-in-noise performance than conventional audiometric thresholds among listeners whose standard hearing falls within the normal range. The findings also point to candidate cutoff values that could eventually help clinicians define what counts as abnormal extended high frequency hearing, a question the field has struggled to settle.

Researchers at the University of South Florida and the University of Central Florida re-analyzed audiological data collected between 2015 and 2024 in two university laboratories. The participants ranged from 18 to 88 years old, with a mean age of about 49 years, and included 246 males and 359 females. Because the Words-in-Noise test used to measure speech-in-noise ability is administered to one ear at a time, the analysis was conducted on 1,205 ears rather than individual participants. Ears were classified using the World Health Organization-aligned criterion for the standard frequency range: 700 ears with a pure tone average of 20 dB HL or less formed the normal hearing group, while 505 ears with averages above 20 dB HL formed the hearing loss group.

The team calculated two pure tone averages for each ear. The standard frequency average covered 0.5, 1, 2, and 4 kilohertz, the frequencies conventionally tested in clinics. The extended high frequency average covered 9, 10, 12.5, and 14 kilohertz, frequencies that lie beyond the reach of a routine audiogram. Speech-in-noise performance was measured with the Words-in-Noise test, in which listeners repeat monosyllabic words presented against multi-talker babble at progressively less favorable signal-to-noise ratios. A higher threshold indicates poorer performance, and the test carries a normative cutoff of 6 dB SNR. A subset of 524 participants also completed the Montreal Cognitive Assessment, a widely used cognitive screener.

Across all 1,205 ears, the two pure tone averages together explained 66.6 percent of the variance in speech-in-noise thresholds, with both showing positive associations: poorer thresholds at either frequency range accompanied worse performance in noise. The picture changed when the analysis was restricted to the normal hearing group. There, the overall model explained 35 percent of the variance in speech-in-noise thresholds, but the extended high frequency average accounted for 17 percent of the variance independently, while the standard frequency average explained only about 1 percent. In other words, among people whose conventional audiograms looked normal, what happened above 8 kilohertz mattered far more for understanding speech in babble.

To probe where abnormal performance begins, the researchers sorted normal hearing ears into 5 dB bins and compared each bin’s speech-in-noise performance against the 6 dB SNR normative criterion. Bins at or below 5 dB HL performed significantly better than the criterion, the bin between 5 and 10 dB HL did not differ significantly from it, and statistically significant impairment emerged beginning in the 10 to 15 dB HL bin and in every worse bin thereafter. This suggests that meaningful speech-in-noise deficits can appear at standard frequency thresholds well below even the most conservative definitions of normal hearing.

The extended high frequency analysis told a parallel story. Impaired speech-in-noise performance was associated with extended high frequency averages as low as 15 dB HL, and the deficits reached statistical significance at averages of 40 dB HL and above, with the abstract reporting significant cutoffs at a standard frequency average above 10 dB HL and an extended high frequency average above 35 dB HL. The contrast between the best and worst performers was striking: among normal hearing participants, the 10 percent with the worst speech-in-noise thresholds had a mean extended high frequency average of 58.7 dB HL, while the best-performing 10 percent averaged just 3.01 dB HL, a difference the model confirmed as highly significant.

Cognition and age entered the analysis as potential confounds, since both are known to relate to hearing and to speech perception. In the 524 participants with cognitive data, a regression model explained about 60 percent of the variance in speech-in-noise thresholds. Extended high frequency sensitivity dominated, explaining 49.3 percent of the variance after removing variance shared with cognitive scores, while the Montreal Cognitive Assessment explained only 1.4 percent independently, with better cognitive scores associated with slightly better performance. When age was added, extended high frequency sensitivity remained the largest predictor, with age and cognition each showing small additional effects. The authors caution that the Montreal Cognitive Assessment is a brief screener rather than a full neurocognitive battery, and that prospective research with more comprehensive cognitive testing is needed before any causal conclusions can be drawn.

The study also surfaced methodological wrinkles with clinical implications. Because data came from two laboratories with different protocols, the Words-in-Noise test was administered at several presentation levels, and the researchers found that sensation level, the loudness of the stimuli relative to a listener’s own thresholds, significantly affected performance. A recent report from the National Academies of Sciences, Engineering, and Medicine endorsed the Words-in-Noise test as a primary clinical outcome measure but did not recommend a standard presentation level, and the authors argue that their sensation level findings suggest those recommendations may be premature. Separately, a multivariate analysis showed that males performed worse than females on speech-in-noise, standard frequency thresholds, extended high frequency thresholds, and cognitive scores.

One puzzling aspect of the results is that the Words-in-Noise stimuli contain limited spectral energy above 8 kilohertz, with the target signal dropping off sharply beyond that point. Yet extended high frequency sensitivity still predicted performance. The authors suggest that damage at the high frequency, basal end of the cochlea may disrupt encoding mechanisms that also serve lower frequency regions classically associated with speech understanding, citing related work linking elevated extended high frequency thresholds to broadened cochlear frequency tuning and to hearing complaints in clinically normal listeners. They note that post-mortem hair cell loss correlates with in-vivo audiograms, supporting the idea that threshold elevation reflects genuine cochlear damage.

The clinical upshot is that extended high frequency audiometry, rarely performed outside research settings, could help explain why some patients with textbook-normal audiograms struggle in restaurants and meetings, a phenomenon sometimes labeled hidden hearing loss. The authors also raise the possibility that such patients, classified as having subclinical hearing loss, might benefit from amplification, though this remains a proposal rather than a tested outcome. The main obstacle they identify is the absence of an accepted cutoff separating normal from abnormal extended high frequency sensitivity. By anchoring that question to functional speech-in-noise performance rather than threshold distributions alone, the study offers a candidate framework, while stopping short of recommending immediate changes to clinical practice until the findings are replicated and a cutoff value is formally established.

Subject of Research: The relationship between extended high frequency hearing sensitivity, cognition, age, and speech-in-noise performance in adults

Article Title: Relationships between extended high frequency hearing, cognition, and speech-in-noise performance

Article References: Dorey, C. M., Eddins, D. A., Eddins, A. C., Higgins, N. C., Secor, C. A., Sanchez, V. A., & Ozmeral, E. J. (2026). Relationships between extended high frequency hearing, cognition, and speech-in-noise performance. Frontiers in Neuroscience, 20, Article 1933436. https://doi.org/10.3389/fnins.2026.1933436

Image Credits: AI Generated

DOI: 10.3389/fnins.2026.1933436

Keywords: audiology, extended high frequency hearing, speech-in-noise, Words-in-Noise test, pure tone average, hidden hearing loss, MoCA, cochlear synaptopathy, hearing loss, cognition, WIN test, normal hearing cutoffs