Source-linked AI summary
Absolute Quality Ratings of Speech Enhancement Systems by Listeners of Different Ages and Degrees of Hearing Loss
Matteo Torcoli, Chih-Wei Wu, Andrea Esposito, Phillip A. Williams, Katrien Cambier, William Wolcott, Antonio Curci, Nicholas S. Reed, Mark Laureyns
TL;DR
Speech-enhancement quality is often evaluated with younger normal-hearing listeners, leaving its relevance to older adults under-explored. This study compared absolute ratings across age and hearing groups using varied dialogue material. Younger listeners’ clear quality differences between systems were smaller or inseparable in older groups, while greater hearing-loss severity was associated with lower ratings.
Problem
How older adults perceive enhanced speech quality, and how system-rating differences from younger normal-hearing listeners translate to them, remains under-explored.
Source: §1 · p. 1Method
The study compared absolute speech-quality ratings from younger normal-hearing listeners and older adults with diverse audiometric profiles using varied, ecologically valid audio material.
Source: §1 · p. 1Results
Takeaways & Limitations
A small older subgroup with mixed audiometric profiles showed younger-like ratings, suggesting audiometry alone cannot explain the contraction in system-rating differences.
Source: §4 · p. 4Future work should test native language and lower output SNRs, use direct-comparison methods, and investigate perceptual priorities, cultural background, subclinical hearing loss, central auditory processing, and cognition.
Source: §4 · p. 4
Abstract
from arXiv · showhide
Speech Enhancement (SE) supports listening, particularly for older adults with age-related hearing loss. Yet, enhanced Speech Quality (SQ) is commonly evaluated by young normal-hearing listeners, and how their ratings translate to older adults remains under-explored. We compared absolute SQ ratings from 40 younger normal-hearing listeners (20-30 years) and 67 older listeners (60-95 years) with diverse audiometric profiles, after screening. Test materials comprised natural dialogues with realistic backgrounds. SQ differences between SE systems that were clear for younger listeners were smaller or inseparable in older groups, regardless of hearing status. Hearing loss severity was associated with lower absolute ratings, but did not strongly modulate the contraction in separable SQ differences. A small, audiometrically mixed subgroup of older listeners showed younger-like rating patterns, suggesting that peripheral audiology alone cannot explain the contraction.
1. INTRODUCTION
Older and younger listeners differ in how they perceive speech quality, but how older adults perceive enhanced speech quality—and how younger listeners’ system ratings translate to them—remains under-explored. This study compares their ratings using varied, ecologically valid audio material.
1. INTRODUCTION
- Speech enhancement is important in technologies used by older adults, yet speech-quality studies and widely used standards commonly rely on younger or normal-hearing listeners.The resulting listener data also inform objective speech-quality metrics.Source: §1 · p. 1
- Speech-quality differences between younger and older listeners are documented, while older adults’ perception of enhanced quality and the transfer of younger listeners’ system ratings remain under-explored.Speech quality can continue improving after intelligibility approaches ceiling, making it a distinct evaluation concern.Source: §1 · p. 1
- This study compares speech-quality ratings across younger normal-hearing listeners and adults aged 60 or older with diverse audiometric profiles, using varied, ecologically valid audio.Source: §1 · p. 1
2. LISTENING TEST
The listening test compared age and hearing groups using dialogue excerpts processed under speech-enhancement conditions and rated for absolute speech quality. It used audiometric screening, controlled playback, and a continuous quality scale.
2.1. Listeners
- After 26 of 133 recruited participants were excluded for test-retest audiometry differences, 107 remained and were grouped as younger normal-hearing and older 60+N0–N3 listeners.The retained groups were defined using reference audiograms and better-ear or, for the sole unilateral-loss participant, worse-ear measurements.Source: §2.1 · pp. 1–2
2.2. Audio Material and Conditions
- The main test used 20 natural dialogue excerpts with realistic backgrounds, each presented under nine conditions including four speech-enhancement systems, three hearing-aid-mode conditions, and two anchors.The systems targeted 20 dB background attenuation; the hearing-aid mode used NAL-R amplification for N3, and the lowest output SNR was 11 dB.Source: §2.2 · p. 2
2.3. Test Methodology
- Listeners rated each clip’s speech quality on a continuous 0–100 scale without direct comparisons, after adjusting playback to a comfortable level and completing pure-tone audiometry before and after rating.The 20 sources were split into sets of 10, yielding 90 main-test items per participant; 86% of listeners’ playback levels were within 4 dB of the calibrated 70 dB SPL default.Source: §2.2 · p. 2, §2.3 · p. 2
3. RESULTS
Across the 60+ groups, condition-rating differences were compressed relative to younger listeners, while ratings also fell with increasing audiometric severity within older listeners. Per-listener modeling showed that severity did not strongly modulate contraction, and language had no significant main effect but a small age-loss interaction.
3.1. Statistical Analysis Framework
- The analysis used Gaussian linear mixed-effects models with listener and source-item random intercepts; logit-transformed scores and source-item split checks left significance decisions unchanged.The primary fit converged without singularity.Source: §3.1 · p. 2
3.2. Overall Condition Ratings by Age-Loss Group
- 21.9 points to 4.4 points separated all four core SE conditions for 30-N0, but the largest contrast was at most 8.4 points in any 60+ group, and the smallest was not significant there.Older listeners showed clear average separation mainly between extreme conditions, whereas younger listeners differed across almost all conditions.Source: Figure 2 · p. 3, §3.2 · p. 3
- SEHA conditions were rated lower than corresponding SE conditions by 30-N0, while the reduction was weak or inconsistent in older groups; the largest 30-N0 SEHA contrast was 5.8 points.Older-group SEHA differences were more uncertain.Source: §3.2 · p. 3
3.3. Modeling Per-Listener Contraction and Lowering
- Condition-averaged EMM ratings were 10.1, 17.9, and 18.4 points lower than 60+N0 for 60+N1, 60+N2, and 60+N3, respectively.Per-listener overall rating level al correlated with PTA (ρ = −0.49, p < 0.001), but not significantly with age or English familiarity.Source: §3.2 · p. 3, §3.3 · pp. 3–4
- All 60+ groups had substantially contracted per-listener slopes; the weak PTA association disappeared when analysis was restricted to the four core SE systems.All older groups retained smaller slopes after logit transformation, so ceiling effects alone did not explain contraction.Source: §3.3 · pp. 3–4
3.4. Language in the Audio Clip
- Language had no significant main effect but a small, significant age-loss interaction; English dubs scored 4.5–7.2 points above paired originals except in 60+N3, where the difference was −0.5.The 60+N3 group was small, so that exception warrants caution.Source: §3.4 · p. 4
4. DISCUSSION
Older listeners showed smaller SQ differences between conditions than younger listeners, alongside lower absolute ratings with greater audiometric severity. The contraction was not fully explained by audiometric category, and fatigue analyses provided no evidence for fatigue or declining concentration as explanations.
4. DISCUSSION
- Older listeners showed contracted SQ differences relative to younger listeners, while absolute ratings fell with increasing audiometric severity.The severity-related lowering was observed within the 60+ cohort, but older listeners with hearing loss sometimes rated conditions higher than younger normal-hearing listeners.Source: §4 · p. 4
- Contraction was also observed for 60+N0, indicating that audiometric category alone cannot fully explain it. Earlier reports attributed the effect to hearing loss, as discussed in, although that study did not observe it.Source: §4 · p. 4
- Trial-half slope changes were negligible and similar across subgroups (−0.04 vs. −0.05; p = 0.65), providing no evidence for fatigue or declining concentration as explanations.Between-listener differences in overall concentration, engagement, or response bias remain possible.Source: §4 · p. 4
- In the exploratory slope-defined split, 5/6 core-condition contrasts were significant in younger-like Subgroup 0, versus 1/6 in Subgroup 1.All 60+ age-loss classes were present in Subgroup 0, suggesting audiometry alone cannot explain the predominant contraction in Subgroup 1.Source: Figure 4 · p. 4, §4 · p. 4
- Screening out listeners with contracted rating profiles could bias age-diverse panels if contraction reflects genuine cohort-specific perception.The authors note that contraction has also been reported in crowdsourced experiments.Source: §4 · p. 4
- Future work should test native language, lower output SNRs, and direct-comparison methods, and investigate perceptual priorities, cultural background, subclinical hearing loss, central auditory processing, and cognition.Source: §4 · p. 4
5. CONCLUSION
For conditions with output SNR ≥ 11 dB, younger listeners’ clear SQ differences between SE systems were smaller or no longer significant in older groups, regardless of hearing status. Greater audiometric severity was associated with lower ratings, while a small mixed-audiometry subgroup showed younger-like ratings.
5. CONCLUSION
- For conditions with output SNR ≥ 11 dB, clear SQ differences between SE systems for younger listeners were smaller or no longer significant in older groups, regardless of hearing status.Greater audiometric severity within the older cohort was associated with lower ratings, and a small subgroup with mixed audiometric profiles showed younger-like ratings.Source: §5 · p. 4