Source-linked AI summary
Optimized Compressor Parameterization for Produced Music in Hearing Aids
Robin Holland, Christoph Hoog Antink, Henning Puder
TL;DR
Hearing-aid compression is typically speech-focused, although music’s dynamics pose distinct fitting questions. This paper maps compression demand across signal and audiogram conditions and optimizes WDRC for selected profiles. Per-signal optimization improves HAAQI over gain-optimized NAL-NL2 baselines while also improving audibility, though subjective validation remains needed.
Problem
Music fitting lacks systematic characterization of compression demand across audiogram shape, residual dynamic range, and signal dynamics, while direct music compressor-parameter optimization remains unexplored.
Source: §1 · p. 1Method
The study characterizes demand across audiometric profiles, residual-DR severities, and produced music signals, then optimizes single- and multi-band WDRC for two profiles.
Source: §1 · p. 1, §2.2 · p. 2Results
Per signal–audiogram optimization improves HAAQI over gain-optimized NAL-NL2 target baselines while also improving audibility.
Source: §5 · p. 4Takeaways & Limitations
Production-compressed recordings fit the residual DR of the evaluated mild-to-moderate profiles under linear gain, while dynamically rich signals and limited residual DR indicate stronger compression demand.
Source: §5 · p. 4The reported gains rely on objective metrics; subjective listening tests are needed to validate them.
Source: §5 · p. 4
Abstract
from arXiv · showhide
Hearing aid compression is typically designed for speech, yet music demands different trade-offs due to its narrower dynamic range (DR) and impulsive transients. This paper investigates whether music fitting benefits from being signal-aware, both in mapping compression demand across signal-audiogram conditions and in tuning its parameters to these combinations rather than inheriting them from speech-derived prescriptions. Compression demand is characterized across eight audiometric profiles, three residual DR severities, and three produced music signals under linear gain. Signal dynamics emerge as a first-order factor: production-compressed music fits the residual DR of the mild-to-moderate profiles evaluated, while dynamically rich recordings challenge even moderate hearing loss. Per-signal optimization of single- and multi-band wide dynamic range compression is then applied to two profiles with the highest compression demand. This optimization improves the Hearing Aid Audio Quality Index over gain-optimized NAL-NL2 baselines while also improving audibility, indicating substantial headroom for music-specific fitting.
1. INTRODUCTION
Hearing-aid research and evaluation remain speech-focused despite music’s distinct dynamics and listening levels. The paper characterizes compression demand across signal and audiogram conditions, then tests per-signal compressor optimization against clinical fitting baselines.
Motivation and research gap
- Music’s narrower dynamic range and higher listening levels create WDRC demands that speech-oriented designs are not optimized to meet.The paper identifies compression-demand variation across hearing-loss profile, residual dynamic range, and signal dynamics as an open question.Source: §1 · p. 1
- Evidence for dedicated commercial music programs is inconclusive, and direct optimization of compressor parameters for music remains unexplored.Speech intelligibility has been optimized through insertion gains and compression thresholds, but not music compressor parameterization.Source: §1 · p. 1
Research questions and contributions
- The study characterizes compression demand across eight hearing-loss profiles and three residual-DR classes, then optimizes 1-, 7-, and 20-channel WDRC for two profiles.Its questions concern demand across signal, audiogram, and residual-DR conditions, and whether per-signal optimization improves objective quality beyond fixed clinical fittings.Source: §1 · p. 1
- Signal dynamics is a first-order factor comparable to residual-DR severity, and multi-band optimization improves HAAQI relative to gain-optimized NAL-NL2 baselines.The reported contribution also identifies a quality–audibility trade-off shaped by audiogram and signal dynamics.Source: §1 · p. 1
2. DATA
The study selects three produced-music clips spanning dynamism and builds representative audiograms from clinical thresholds and uncomfortable loudness levels, stratified by residual dynamic range.
2.1. Music Signal Selection
- Three Free Music Archive tracks represent Tight, Normal, and Dynamic signals after genre, popularity, and manual production-style screening.Tracks are characterized by spectral macrodynamics and transient microdynamics; the final 20 seconds of each clip are evaluated.Source: §2.1 · pp. 1–2
2.2. Audiogram Dataset
- Audiograms come from OHHR measurements and are classified into standard profiles, then stratified into High, Moderate, and Limited residual-DR groups.N6 and N7 are excluded for insufficient data; each group is represented by the medoid in joint threshold–UCL space.Source: §2.2 · p. 2
3. ANALYSIS
The analysis defines bandwise compression effectiveness around audibility and UCL compliance, maps where linear gain may be insufficient, and specifies per-signal WDRC optimization and its comparison baselines.
3.1. Compression Effectiveness Metric
- Compression effectiveness assesses whether amplified music stays within usable hearing range, tracking sustained levels and brief transient peaks separately.The slow path uses 400-ms windows with 100-ms hops, while the fast path uses 20-ms windows with 5-ms hops; Fig. 1 maps GUE and GLE by HL profile and residual-DR group for three signals.Source: §3.1 · p. 2, Figure 1 · p. 3
3.2. Compression Demand Under Linear Gain
- NAL-R linear gain is applied across signal and audiogram–residual-DR conditions, with broadband gain optimized to isolate compression demand.Conditions with GUE or GLE substantially below 1.0 are candidates for WDRC.Source: §3.2 · p. 2
3.3. Compressor Parameter Optimization
- For two limited-residual-DR profiles, the study optimizes 1-, 7-, and 20-channel WDRC while holding NAL-NL2 insertion gain, T1, and attack and release times fixed.The free static-characteristic parameters are T2, R1, and R2; multi-band parameters are optimized at anchor channels and interpolated.Source: §3.3 · p. 2
- The optimization balances HAAQI, audibility, and UCL compliance, and compares optimized models with NAL-NL2 and NAL-R target baselines.The objective uses fixed weights, evaluates trials at 55 and 90 dB SPL, and scores HAAQI against a NAL-R linear-gain reference.Source: §3.3 · pp. 2–3
4. RESULTS & DISCUSSION
Compression demand depends strongly on signal dynamics as well as hearing-loss severity and residual dynamic range. Per-signal WDRC optimization improves HAAQI over gain-optimized NAL-NL2 baselines, while quality and audibility trade-offs vary with channel count and audiogram shape.
4.1. Compression Demand Under Linear Gain
- For the worst cases—N4 and N5 with limited DR and the Dynamic signal—GUE is 0.86 and GLE is 0.75.Compression effectiveness degrades with increasing hearing-loss severity and decreasing residual DR.Source: §4.1 · p. 3
- Across signals, GLE spread reaches 0.15 for a given audiogram, comparable to residual-DR severity; GUE varies by only 0.04.The authors suggest production mastering compresses peaks similarly across signals, while quiet-passage levels remain genre-dependent.Source: §4.1 · p. 3
4.2. Optimization Results
- All optimized WDRC configurations achieve GUE ≥0.99 and TGUE ≥0.98, whereas NAL-R has TGUE of 0.91–0.97.N4 and S2 with limited residual DR were selected for optimization because each had GUE <0.93 and GLE <0.87 for at least one signal.Source: §4.2 · p. 3, §4.1 · p. 3
- Per-signal optimization improves HAAQI by 0.12 on average (range 0.03–0.22) over gain-optimized NAL-NL2 baselines.The comparison covers N4 and S2 with limited residual DR, averaged over 55 and 90 dB SPL.Source: §4.2 · pp. 3–4, Table 1 · p. 4
- WDRC-1 has the highest HAAQI, but WDRC-20 raises GLE over WDRC-1 by 0.06 on average for sloping S2, with values 0.92, 0.94, and 0.84.For flat N4, multi-band models improve GLE over WDRC-1 only for the Dynamic signal (+0.09); no configuration dominates both axes.Source: §4.2 · p. 4, Figure 3 · p. 4
5. CONCLUSION
Per signal–audiogram optimization improves HAAQI over gain-optimized NAL-NL2 baselines while also improving audibility. Music dynamics and residual DR shape compression demand, supporting signal-aware fitting rather than fixed audiogram-based configurations.
5. CONCLUSION
- The paper assesses compression demand and tunes compression parameters to the signal–audiogram combination when linear gain is insufficient.Source: §5 · p. 4
- Music dynamics rival residual DR severity as a determinant of compression demand; production-compressed recordings fit evaluated mild-to-moderate profiles under linear gain.The strongest case for compression is in dynamically rich signals and/or with limited residual DR, motivating signal–audiogram-adaptive decisions rather than fixed audiogram-based music programs.Source: §5 · p. 4
- Per signal–audiogram optimization improves HAAQI over gain-optimized NAL-NL2 (G2) target baselines while also improving audibility.Maximal granularity over the static compression characteristic establishes an achievable ceiling and quantifies headroom for less granular strategies; gain adjustment and compression tuning serve largely orthogonal roles. The benefit of multi-band compression depends on audiogram shape, with the sloping profile gaining from finer frequency resolution.Source: §5 · p. 4
- Results rely on objective metrics, making subjective listening tests the essential next step to validate the observed gains.The authors also identify broader signal and audiometric coverage as needed to support deployable signal-adaptive fitting strategies.Source: §5 · p. 4