Organizations: 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 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.
Figures & tables
Figure 1: Example questions from the Voices of Civilizations benchmark on three folk songs—Arabic "Jafra," Chinese "Liuyang River", and Korean "Arirang."
Figure 2: Per-language accuracy (%) of three state-of-the-art audio LLMs on the VoC benchmark using audio input only and focusing on region, mood, and theme questions. We invited a Chinese music teacher to answer 29 questions across 10 Chinese songs in a strictly closed-book setting (no reference or lookup allowed), achieving an accuracy of 79.31%.
Setting
Language
Region
Mood
Theme
Noise
Gemini 2.5 Pro
93.42
42.48
40.15
47.50
Qwen2.5-Omni-7B
51.05
26.11
23.48
31.25
Kimi-Audio-7B-Instruct
40.26
23.11
23.45
24.06
Audio (Eng QA)
Gemini 2.5 Pro
99.74
73.01
62.50
85.00
Table 1: Accuracy (%) of three audio LLMs on four question types in the VoC benchmark under different settings.
The evaluation of music understanding in Large Audio-Language Models (LALMs) requires a rigorously defined benchmark that truly tests whether models can perceive and interpret music, a standard that current data methodologies frequently fail to meet. This paper introduces a meticulously structured approach to music evaluation, proposing a new dataset of 320 hand-written questions curated and validated by experts with musical training, arguing that such focused, manual curation is superior for probing complex audio comprehension. To demonstrate the use of the dataset, we benchmark six state-of-the-art LALMs and additionally test their robustness to uni-modal shortcuts.
Benno Weck, Pablo Puentes, Andrea Poltronieri +2
Universitat Pompeu Fabra · Universitat Autònoma de Barcelona
Multilingual fluency often invites a stronger assumption: a model that can speak a user's language must also understand the culture encoded by that language. We call this the Illusion of Cultural Alignment. To test this assumption directly, we introduce MSQA, a benchmark of 1,064 natively sourced questions across 11 language groups, five cultural dimensions, and three difficulty tiers. Unlike translated benchmarks, MSQA targets locally grounded knowledge and reduces shortcuts from English-centric cross-lingual transfer. Evaluating 18 LLMs, we find substantial cultural degradation and a pronounced Locality Effect: cultural competence tracks pre-training exposure more closely than general reasoning ability. We further show that common inference-time remedies do not dissolve the illusion. Models remain overconfident on unfamiliar cultural questions, repeated sampling yields unstable rather than reliable correctness, and retrieval augmentation helps unevenly on long-tail facts. These findings indicate that cultural alignment cannot be inferred from multilingual ability alone and requires deeper intervention than calibration, sampling, or retrieval at inference time
Recent work on music question answering (Music-QA) has primarily focused on single-track understanding, where models answer questions about an individual audio clip using its tags, captions, or metadata. However, listeners often describe music in comparative terms, and existing benchmarks do not systematically evaluate reasoning across multiple tracks. Building on the Jamendo-QA dataset, we introduce Jamendo-MT-QA, a dataset and benchmark for multi-track comparative question answering. From Creative Commons-licensed tracks on Jamendo, we construct 36,519 comparative QA items over 12,173 track pairs, with each pair yielding three question types: yes/no, short-answer, and sentence-level questions. We describe an LLM-assisted pipeline for generating and filtering comparative questions, and benchmark representative audio-language models using both automatic metrics and LLM-as-a-Judge evaluation.