Modern audio systems universally employ mel-scale representations derived from 1940s Western psychoacoustic studies, potentially encoding cultural biases that create systematic performance disparities. We present a comprehensive evaluation of cross-cultural bias in audio front-ends, comparing mel-scale features with learnable alternatives (LEAF, SincNet) and psychoacoustic variants (ERB, Bark, CQT) across speech recognition (11 languages), music analysis (6 collections), and European acoustic scene classification (10 European cities). Our controlled experiments isolate front-end contributions while holding architecture and training protocols minimal and constant. Results demonstrate that mel-scale features yield 31.2% WER for tonal languages compared to 18.7% for non-tonal languages (12.5% gap), and show 15.7% F1 degradation between Western and non-Western music. Alternative representations significantly reduce these disparities: LEAF reduces the speech gap by 34% through adaptive frequency allocation, CQT achieves 52% reduction in music performance gaps, and ERB-scale filtering cuts disparities by 31% with only 1% computational overhead. We also release FairAudioBench, enabling cross-cultural evaluation, and demonstrate that adaptive frequency decomposition offers practical paths toward equitable audio processing. These findings reveal how foundational signal processing choices propagate bias, providing crucial guidance for developing inclusive audio systems.
Figures & tables
Frequency
Mel Value
Bandwidth
Required JND
Deficit
(Hz)
(Hz)
(Hz)
Ratio
80
122.0
51.6
0.8
65×
100
150.5
51.6
1.0
52×
200
283.2
58.6
2.0
29×
250
344.2
62.4
2.5
25×
300
402.0
66.4
3.0
22×
Table 1: Mel-scale resolution deficit across the 80–500 Hz range critical for tonal distinctions, based on a standard 40-filter filterbank (0–8000 Hz). Bandwidth is the center spacing in Hz; JND is approximately 1% of frequency.
Figure 1: Performance gaps across domains. Mel baseline shows 12.5% speech gap and 15.7% music gap. Percentage improvements are shown above bars.
Speech (WER/CER %)
Music (F1 %)
Scenes (Acc %)
Comp.
Front-end
Tonal
Non-tonal
Non-West
West
Europe-1
Europe-2
Cost (Overhead)
mel
31.2 ± 1.2
18.7 ± 0.8
56.7 ± 2.1
72.4 ± 1.5
71.2 ± 1.4
76.8 ± 1.2
1.00 × (0%)
ERB
26.4 ± 1.0
17.8 ± 0.7
62.8 ± 2.0
73.1 ± 1.4
72.6 ± 1.3
77.2 ± 1.1
1.01 × (1%)
Bark
27.2 ± 1.0
18.1 ± 0.8
61.9 ± 2.1
72.8 ± 1.5
72.2 ± 1.3
76.9 ± 1.2
1.01 × (1%)
CQT
28.8 ± 1.1
19.2 ± 0.9
65.3 ± 1.9
72.9 ± 1.4
–
–
1.15 × (15%)
LEAF
25.8 ± 0.9
17.5 ± 0.7
62.4 ± 2.0
73.5 ± 1.4
72.5 ± 1.3
77.5 ± 1.1
1.08 × (8%)
Table 2: Comprehensive performance metrics across domains. Bottom rows show fairness metrics (WGS, Δ , ρ ) with baseline → best achieved values.
Figure 2: LEAF’s learned frequency allocation. Tonal languages allocate 42% of filters to the critical 80-500 Hz range (vs 23% for mel).
ASR systems exhibit persistent performance disparities across accents, but whether these gaps reflect superficial biases or deep structural vulnerabilities remains unclear. We introduce ACES, a three-stage audit that extracts accent-discriminative subspaces from ASR representations, constrains adversarial attacks to them, and tests whether removing them improves fairness. On Wav2Vec2-base with seven accents, imperceptible perturbations (~60 dB SNR) along the accent subspace amplify the WER disparity gap by nearly 50% (21.3->31.8 pp), exceeding random-subspace controls; a permuted-label test confirms specificity to genuine accent structure. Partially removing the subspace worsens both WER and disparity, revealing that accent-discriminative and recognition-critical features are deeply entangled. ACES thus positions accent subspaces as powerful fairness-auditing tools, not simple erasure levers.
We introduce Voices of Civilizations, the first multilingual QA benchmark for evaluating audio LLMs' cultural comprehension on full-length music recordings. Covering 380 tracks across 38 languages, our automated pipeline yields 1,190 multiple-choice questions through four stages - each followed by manual verification: 1) compiling a representative music list; 2) generating cultural-background documents for each sample in the music list via LLMs; 3) extracting key attributes from those documents; and 4) constructing multiple-choice questions probing language, region associations, mood, and thematic content. We evaluate models under four conditions and report per-language accuracy. Our findings demonstrate that even state-of-the-art audio LLMs struggle to capture subtle cultural nuances without rich textual context and exhibit systematic biases in interpreting music from different cultural traditions. The dataset is publicly available on Hugging Face to foster culturally inclusive music understanding research.
Shangda Wu, Ziya Zhou, Yongyi Zang +4
Independent Researcher · The Hong Kong University of Science and Technology, Hong Kong SAR, China · The Chinese University of Hong Kong, Hong Kong SAR, China
We introduce HEAR (Human-recorded Evaluation of Audio-LLM bias by Real speakers), a large-scale, ecologically valid benchmark comprising 87k real human audio samples from 843 demographically diverse participants. HEAR enables comprehensive evaluation through Multiple Choice Question Answering (MCQA) and open-ended long-form tasks. To our knowledge, this is the first large-scale voice benchmark grounded entirely in authentic human speech. We evaluate model behavior across both real-time speech-to-speech and speech-to-text architectures. Our results reveal that voice-conditioned bias is a model-specific property. Furthermore, we demonstrate that personalization instructions consistently exacerbate demographic disparities. Our findings establish that voice bias is a controllable model characteristic, providing a foundational framework for future bias mitigation and evaluation in Audio-LLM development.