每日 arXiv

音频 · 语音 · 音乐 · 声学

每日新爬取的 arXiv 论文，按评分排序。中文摘要帮你快速判断是否值得深读。

2026-07-11
日期2026-07-11
已评分—
均分—
最高—

Daily Papers — 2026-07-11

17 papers on audio, speech, music, and acoustics.

1. Hearing Like Humans? Sound Symbolism and Perceptual Alignment in Speech Language Models

Authors: Yun-Shao Tsai, Chun-Wei Chen, Chee-En Yu, Yi-Cheng Lin, Hung-yi Lee

Categories: eess.AS, cs.CL | Submitted to SLT 2026 Score: 7.40/10 (Obj:8 Id:8 Ind:8 Comp:8 Eff:6 Nov:8 Rep:7)

  • Strength: 首次使用真实人类语音录音系统评估8个SLM的声音象征性，通过四实验分解精确定位对齐缺陷在听觉表示而非视觉感知（视觉r=.94-.97 vs 听觉r=.470 max），并经RSA揭示模型未使用人类依赖的spectral tilt线索（人类ρ=.431 vs 模型最大.057）。
  • Weakness: 核心结论”语音表示需捕获人类声学线索”过于宽泛缺乏可操作化建议，且Gemini3.5-Flash的完美匹配可能源于词汇记忆而非声学处理，这一关键替代解释未经非词化刺激实验排除。
  • Full review: Claude Code 全文七公理审稿

Sound symbolism, the human tendency to map speech sounds to perceptual qualities such as roundness or sharpness, arises primarily from the acoustics of speech rather than spelling. Whether Speech Language Models (SLMs) share this tendency remains open, as prior evaluations rely on text or images rather than real speech. We study it using genuine human speech recordings, comparing model judgments against human data across the auditory, crossmodal, and visual components of the effect. We find that SLMs’ auditory judgments align poorly with human perception and miss the acoustic cues, such as spectral tilt, that drive human intuitions, and open-weight models cannot reliably link a heard sound to its corresponding shape. With a visual-only control ruling out shape perception, the weakness localizes to how speech is represented, suggesting that perceptual alignment depends not on stronger vision but on speech representations that capture the cues humans hear.

2. Learn2Chat: Rethinking Dyadic Talking Heads via Interaction-Modulated Monologic Priors

Authors: Zikai Huang, Siyue Chen, Xuemiao Xu, Haoxin Yang, Cheng Xu et al.

Categories: cs.GR, cs.SD Score: 6.74/10 (Obj:8 Id:5 Ind:8 Comp:5 Eff:8 Nov:5 Rep:4)

  • Strength: 在 DualTalk test/OOD 上全面取得 SOTA，FD 降低 ~26%，rPCC 降低 ~16%，参数仅 48M（DualTalk 638M 的 8%），且验证了 plug-and-play 兼容两种 backbone。
  • Weakness: 缺少“直接微调独白模型”这一最简 baseline 的识别证据；未引用/讨论 CVPR 2025 同领域重要工作 INFP；代码和超参数未公开；方法由 5 个损失项+2 阶段训练构成的复杂 pipeline 缺乏对更简替代方案的系统比较。
  • Full review: Claude Code 全文七公理审稿

Dyadic conversational motion generation is essential for realistic interactive digital humans. Existing approaches typically model conversational behaviors within unified dyadic generators. However, such holistic formulations tend to couple self-speech-driven motion with partner-responsive social feedback, leaving the interaction-specific component implicit and underutilizing the speech-motion correspondence already learned by pretrained monologic motion models. We propose Learn2Chat, a unified framework that models dyadic motion as interaction modulation over pretrained monologic motion priors. This design separates intrinsic speech-driven motion from social interaction effects and enables more structured interaction modeling. Specifically, we introduce a Monologic-Anchored Motion Factorization scheme that leverages the semantic motion manifold learned from monologic data to disentangle audio-driven motion dynamics from interaction-induced modulation, yielding clean interaction representations from dyadic sequences. On top of this representation space, a Cross-Attentive Interaction Latent Prediction module maps paired speech signals to interaction latents through cross-branch attention and interaction alignment. During inference, the predicted interaction latents modulate canonical monologic motion to generate coherent and synchronized dyadic behaviors in a data-efficient manner. Extensive experiments on the DualTalk benchmark demonstrate that Learn2Chat achieves state-of-the-art performance across both quantitative metrics and perceptual evaluations. Moreover, the framework is model-agnostic and seamlessly integrates with diverse pretrained monologic motion backbones, highlighting the effectiveness of prior reuse and interaction adaptation for scalable conversational motion generation. More visual results are available on the project page.

3. Partial Contracts Suffice: Sound, LLM-Inferred Regression Verification

Authors: Yiannis Charalambous, Rafael Menezes, Youcheng Sun, Lucas C. Cordeiro

Categories: cs.SE, cs.AI | 12 pages Score: 6.60/10 (Obj:8 Id:6 Ind:8 Comp:8 Eff:6 Nov:8 Rep:5)

  • Strength: 定理 T1 形式化证明了单边部分契约对安全保留条件回归等价性是可靠的，且 RQ2 显示该部分契约在可比较目标中已达到近乎完全的紧密度（18/20, 13/15, 14/18 等价于强化版本），验证了部分规范停止点的合理性。
  • Weakness: EqBench-C 上仅决定 25.7% 的配对（70/272），即使加入契约也仅 38.6%，且恢复决策中约 38% 为误报；全部基准为小规模（~17 LoC 或六个函数），缺少生产代码或跨工具公平重跑对比。
  • Full review: Claude Code 全文七公理审稿

Software evolves continuously, yet ensuring that a patch preserves intended behavior without re-verifying an entire codebase remains difficult. Regression verification addresses this problem, but existing techniques require expensive whole-program reasoning or rely on manually written specifications that are rarely available in practice. We present the first contract-based regression verification tool. Contract soundness is ensured by proving all function versions match the behavior. The contract then verifies program flow via assume-guarantee. We ask whether a partial, caller-sufficient contract, rather than a full behavioral specification, is enough. On Frama-C-Problems we strengthen each inferred contract past what the caller needs and measure how much tighter it becomes. It barely moves: for most targets in every model the caller-sufficient contract is already the tightest the loop reaches, and our tightness comparator rates the partial and strengthened contracts equivalent for the large majority of targets it can compare. Partial-spec contracts thus capture nearly all the attainable tightness, so stopping at caller-sufficiency costs almost nothing. The regression check underneath is sound: on the third-party EqBench-C suite it never fabricates an equivalence, returning zero false proofs and reporting an unprovable difference instead. It also surfaced nine pairs that EqBench mislabels as equivalent, more than a concurrent tool reports. The contracts themselves are inferred automatically from the checker’s own counterexamples, with no separate specification step; on Frama-C-Problems and the ANSSI X509 parser this reaches a verification rate comparable to tools AutoSpec and Preguss, while a passing result certifies at least as strong a property, which we call \emph{safety-preserving conditional equivalence}: enforcement plus caller-sufficiency.

4. CoFi-Lite: Pushing the Limits of Ultra-Lightweight Speech Enhancement

Authors: Leyan Yang, Dahan Wang, Xiaobin Rong, Jiadong Zhao, Jing Lu

Categories: eess.AS | Accepted by IEEE Signal Processing Letters Score: 6.32/10 (Obj:8 Id:6 Ind:5 Comp:5 Eff:8 Nov:5 Rep:7)

  • Strength: 在 12.87M MACs/s 和 83.12k 参数的极低复杂度下，通过双路径 coarse/fine 并行解耦 + CPF 模块，在 DNS3 上达到 PESQ 2.16，全面超越 Level I 基线且 MACs/s 最低。
  • Weakness: 4 个 baseline 中 3 个（GTCRN, UL-UNAS, AdaptCRN）来自同一实验室且部分作者重叠，baseline 比较的独立性存疑；各架构组件多为已有模块重组，缺乏对方法有效机制的深层解释。
  • Full review: Claude Code 全文七公理审稿

Ultra-lightweight models are essential for the deployment of deep learning-based speech enhancement algorithms on edge devices. Although recent approaches have achieved a certain balance between computational complexity and performance, pushing the complexity limits further demands more sophisticated designs. In this letter, we propose CoFi-Lite, a highly efficient model that decouples spectral modeling into coarse- and fine-grained streams. By leveraging two parallel and symmetric encoder-decoder paths, it simultaneously extracts full-band envelopes and low-frequency details for complementary enhancement. In addition, a novel Cross-Path Fusion (CPF) module is introduced to bridge the distinct paths, facilitating efficient feature interaction. Remarkably, CoFi-Lite requires extremely low computational resources, featuring only 12.87M MACs/s and 83.12k parameters. Experimental results demonstrate that our proposed model outperforms the ultra-lightweight baseline GTCRN while requiring only 40.26% of its computational complexity. Its scaled-up variant also delivers performance on par with that of the SOTA ultra-lightweight model AdaptCRN alongside a 19.34% reduction in computational cost. Audio examples are available at https://acceleration123.github.io/CoFiLite-demo/.

5. GigaAM Multilingual: Foundation Model for Underrepresented Languages

Authors: Andrei Kuzmenko, Alexandr Maximenko, Aleksandr Kutsakov, Georgii Gospodinov, Dmitrii Bolotov et al.

Categories: eess.AS, cs.CL | Accepted to Interspeech 2026. Model weights: https://github.com/salute-developers/GigaAM Score: 6.30/10 (Obj:8 Id:6 Ind:5 Comp:5 Eff:8 Nov:5 Rep:7)

  • Strength: 在 Kazakh/Kyrgyz/Uzbek 上实现大幅 WER 降低（如 Kyrgyz CV 10.2% vs Whisper 95.2%），且 240M 参数模型在 matched CTC fine-tuning 下即超越 Whisper Large v3 和 Omnilingual-1B，证明 encoder 预训练策略的高效性。
  • Weakness: 核心 method 增量（cluster-level balancing, domain-aware sampling）缺少与最简直接替代方案（language-level balancing）的对比消融，且各组件均已有直接先行工作（mHuBERT-147, GigaSpeech 2, 原始 GigaAM），方法论新颖性有限。
  • Full review: Claude Code 全文七公理审稿

Despite recent scaling successes, multilingual ASR performance remains highly uneven, with long-tail languages suffering from severe data scarcity. This work addresses the challenge of building robust foundation models for underrepresented Central Asian languages (Kazakh, Kyrgyz, Uzbek). We present GigaAM Multilingual, a Conformer encoder pre-trained on 2M hours of audio using a HuBERT-style objective. Crucially, we introduce a cluster-level data balancing strategy during pre-training and a domain-aware sampling method during fine-tuning to mitigate head-language dominance. In controlled comparisons, our approach outperforms strong open pretrained encoders (Whisper Large v3, Omnilingual-1B) on target languages, achieving significant gains on spontaneous speech while maintaining efficiency. We release the foundation encoder and ASR model, offering a proven recipe for effective multilingual adaptation under realistic data imbalance.

6. GigaChat Audio: Time-aware Large Audio Language Model

Authors: Aleksandr Kutsakov, Mariia Sadovina, Georgii Gospodinov, Alexandr Maximenko, Oleg Kutuzov et al.

Categories: eess.AS, cs.CL | Accepted to Interspeech 2026. Model and dataset: https://huggingface.co/ai-sage/GigaChat3.1-Audio-10B-A1.8B Score: 6.21/10 (Obj:8 Id:6 Ind:6 Comp:8 Eff:5 Nov:5 Rep:8)

  • Strength: 在 120 分钟长音频 temporal grounding 上实现 53.8–65.2 mIoU，远超 Qwen3-Omni（3.6），并通过系统性消融证明周期性时间锚点的必要性和稀疏锚点（1/min）的充分性，开源权重和数据集。
  • Weakness: 在公开短音频 benchmark 上全面落后于 TimeAudio 和 Gemini 3 Flash，跨模型比较不公平（10B vs 30B vs 闭源），且未引用高度相关的同期工作 “Listening with Time”（2604.22245），核心 idea 从 Video-LLM 迁移而非原创。
  • Full review: Claude Code 全文七公理审稿

Temporal grounding in long recordings remains challenging for audio-conditioned LLMs. We present a time-aware audio LLM that answers questions with explicit timestamps over up to 120 minutes of input. Our approach interleaves periodic time markers with continuous audio tokens using large-scale synthetic supervision from a cascaded pipeline. Our model achieves strong temporal-grounding accuracy on short and long benchmarks and supports time-anchored fragment descriptions and summaries. Extensive ablations examine how time representation, marker frequency, tokenization, and duration-mixture design affect accuracy and computational cost. We release model weights and datasets to support further research on time-aware audio understanding, available at https://huggingface.co/ai-sage/GigaChat3.1-Audio-10B-A1.8B.

7. FdAudio: MeanFlow-Anchored Fréchet-Distance Post-Training for One-Step Text-to-Audio Generation

Authors: Kuan-Po Huang, Bo-Ru Lu, Ho-Lam Chung, Shih-Hsin Wang, Hung-yi Lee

Categories: eess.AS, cs.SD | Project website: https://fdoneaudio.github.io/ Score: 5.90/10 (Obj:8 Id:6 Ind:5 Comp:5 Eff:7 Nov:5 Rep:8)

  • Strength: FdAudio 通过 FD-loss post-training + MeanFlow anchor 在 AudioCaps 上实现 few-step T2A SOTA，1-step FD=12.71/FAD=1.26（较 MeanAudio 分别降低 11.4%/28.8%），同时保持 25-step 轨迹不退化（Table III 消融证实 anchor 必要性）。
  • Weakness: 核心方法是 FD-loss（2604.28190）与 MeanFlow（2505.13447）的直接组合，FD-loss 本身缺乏单独消融，仅 AudioCaps 单数据集验证，且 model selection 以 FD 为标准与核心评测指标构成轻度循环。
  • Full review: Claude Code 全文七公理审稿

While recent few-step sampling text-to-audio generation models like MeanAudio substantially accelerate generation by modeling average velocities, their strict one-step generation quality still lags significantly behind multi-step counterparts. We propose FdAudio to bridge this gap. Unlike MeanAudio, which relies solely on regression against target velocity fields, our post-training approach optimizes the final one-step distribution directly across pre-trained embedding spaces via a multi-representation Fréchet-distance (FD) loss. Crucially, to prevent the multi-step degradation that naive post-training with FD-loss causes, we introduce a MeanFlow consistency objective as a structural anchor. Results demonstrate that FdAudio establishes state-of-the-art one-step T2A generation quality among few-step systems, yielding an 11.4% reduction in FD score and a 28.8% improvement in FAD score relative to the baseline MeanAudio framework. Notably, we solve FD post-training’s naive multi-step degradation issue by proposing the MeanFlow anchor, enabling a 25-step sampling path to maintain high-fidelity audio synthesis that matches or surpasses strong multi-step models at a fraction of their computational latency.

8. Breaking the Quality–Intelligibility Trade-off in Streaming Target Speaker Extraction via Deep-Feature-Anchored Preference Optimization

Authors: Shuhai Peng, Jinjiang Liu, Hui Lu, Liyang Chen, Guiping Zhong et al.

Categories: cs.SD, cs.AI Score: 5.80/10 (Obj:8 Id:6 Ind:5 Comp:5 Eff:5 Nov:5 Rep:4)

  • Strength: 通过对照三种 DPO 排序准则（DNSMOS/WER/WavLM），诊断出感知指标优化的 reward hacking 机制（辅音瞬态被系统性擦除），并发现 WavLM 锚定 DPO 可实现 WER 10.9% 相对改善（0.138→0.123）同时保持质量与说话人相似度不退化。
  • Weakness: DPO 仅贡献 4.7% 边际改善且最优结果依赖 checkpoint selection（e0=0.123, e2=0.138 缺乏稳定性），训练锚点 WavLM 与评测指标 WavLM Sim 重叠构成循环风险，SIG 退化 0.05 被淡化，且缺少 MeanFlow-TSE/FlowTSE/TargetVoice 等同期 TSE 方法的 baseline 对比。
  • Full review: Claude Code 全文七公理审稿

Generative streaming models for Target Speaker Extraction (TSE) commonly exhibit a quality–intelligibility trade-off, wherein naive optimization for perceptual audio quality tends to degrade speech intelligibility, and conversely. We reveal that this trade-off arises not from the constraints of streaming architectures, but from an inappropriate choice of optimization anchor. Directly optimizing against audio quality metrics induces catastrophic reward hacking, where content critical to pronunciation and intelligibility is systematically erased to maximize a proxy score. To break this bottleneck, we propose two complementary improvements: an enlarged Conformer convolution kernel for richer local spectro-temporal modeling, and WavLM-anchored Direct Preference Optimization (DPO) fine-tuning strategy. DPO preference pairs are ranked by WavLM cosine similarity, a deep acoustic feature encoding both phonetic structure and speaker identity, providing an optimization anchor that resists hacking. Under a 560 ms streaming chunk size, the proposed method achieves a 10.9% relative intelligibility improvement (word error rate: 0.138 to 0.123), with marginal simultaneous gains in audio quality and speaker similarity.

9. WaveNet-Style Guitar Amplifier Model Pruning for Real-Time iOS Deployment

Authors: Ryota Sato, Eli Silverstein

Categories: eess.AS, cs.SD, eess.SP | Accepted to DAFx 2026 Demo Score: 5.70/10 (Obj:8 Id:5 Ind:8 Comp:4 Eff:6 Nov:5 Rep:8)

  • Strength: 90% 迭代剪枝 + 自研稀疏 C++ 引擎在 iPhone 16 Pro 上实现 RTF ≈ 0.6 的实时 WaveNet 放大器仿真，ESR < 3.4e-4，代码开源。
  • Weakness: 方法为已有技术（WaveNet + 迭代幅度剪枝 + 稀疏推理）的工程组合，无新机制；缺少与 Südholt et al. 2023 等直接同类工作的定量对比，”no perceptual loss”仅基于 informal listening 无正式感知评估。
  • Full review: Claude Code 全文七公理审稿

WaveNet-style convolutional networks emulate tube amplifiers and distortion pedals with high fidelity, but their computational cost has confined them to desktops or dedicated DSP hardware. We present a sparse-enabled WaveNet inference engine for iOS that runs heavily pruned neural guitar amplifier models in real time on iPhones. Aggressive iterative magnitude pruning removes 90% of the network weights with no perceptible loss in quality. A custom sparse C++ engine turns this sparsity directly into compute savings, sustaining low-latency real-time operation on a CPU-only iPhone implementation where the dense model cannot. On-device output matches the trained model to within int16 quantization error. At the demonstration, visitors will play a guitar through the app on iPhone hardware and A/B the on-device pruned model against the physical pedal it emulates. Source code and audio examples are available at https://github.com/ryos17/wavenet-imp.

10. MeloBottleneck: Self-Supervised Melody Skeleton Extraction with a Latent Subsequence Bottleneck

Authors: Fan Bu, Rongfeng Li, Linfeng Fan

Categories: cs.SD, cs.LG | 8 pages, 3 figures Score: 5.60/10 (Obj:6 Id:5 Ind:5 Comp:6 Eff:5 Nov:5 Rep:8)

  • Strength: 将旋律骨架提取重构为长度可控的保序潜在子序列 + rhythmic closure，自监督训练在跨域 V2T 上超越所有 baseline（Hard F1 0.6677），并显著提升 BM25 检索（R@1 从 0.117 提升至 0.208）。
  • Weakness: 在合成 O2B 上输给 matched pseudo-label classifier（0.8809 vs 0.8942），在 O2G 上输给 Top-K Duration 最简 heuristic（0.8942 vs 0.9317），且 V2T 评测目标为 alignment-induced 而非人工标注，O2G 仅 20 条序列，评测独立性与统计效力不足。
  • Full review: Claude Code 全文七公理审稿

Melody skeleton extraction aims to derive a shorter melody that preserves structural notes while removing ornaments. Prior methods rely on hand-crafted reduction rules or note-wise salience classifiers trained with heuristically or procedurally generated pseudo-labels. Such supervision can inherit generator bias and does not explicitly optimize a coherent reduced melody. We introduce MeloBottleneck, a self-supervised framework that represents a skeleton as a length-controlled, order-preserving latent subsequence. A hard-bottleneck extractor selects note events, a rhythmic-closure operator produces a self-consistent skeleton, and a re-ornamentation decoder reconstructs the input melody. Training combines reconstruction, a frozen autoregressive melody prior, ornament-invariant consistency across procedurally ornamented views, and ornament exclusion. We evaluate three regimes: synthetic out-of-distribution ornament-to-skeleton, TAVERN variation-to-theme, and Jiugong ornamented-to-gongche. A matched pseudo-label classifier excels on the synthetic benchmark, while MeloBottleneck transfers better, achieving competitive selection quality on TAVERN and Jiugong. Skeletonized melodies also improve BM25-based fragment retrieval, boosting Recall@K and MRR while reducing query time. Overall, the results suggest that learning skeletons as latent subsequences yields more robust transfer than pseudo-label imitation.

11. Perceived Annoyance in Multi-source Electric Vehicle AVAS Environments

Authors: Berkay Kullukcu, Jonas Krautwurm, Serkan Atamer, Ercan Altinsoy

Categories: eess.AS, eess.SP Score: 5.60/10 (Obj:8 Id:5 Ind:8 Comp:5 Eff:5 Nov:5 Rep:4)

  • Strength: 首次在受控双耳实验中验证多源 EV AVAS pass-by 场景的短期烦恼非加性——同时双车场景烦恼较单车显著增加 6.54 分（p=0.012, Holm 校正后 p=0.036），且 LAeq 无法解释该增加，支持 scene-based 评估框架。
  • Weakness: N=10 样本量过小，核心显著性仅在”同时通过”子条件存活，时间偏移条件和合并检验均在校正后失去显著性，且 psychoacoustic descriptor 分析不完整（仅报告 impulsiveness，其他指标未报告），数据/代码未开源。
  • Full review: Claude Code 全文七公理审稿

The increasing usage of electric vehicles in urban environments has resulted in a widespread presence of AVAS sounds. While individual vehicle sound design and testing is a common approach, real-world traffic scenarios often involve the simultaneous presence of multiple vehicles. Their combined presence may lead to changes in perception, compared to when they are presented individually, specifically regarding annoyance. The work addresses annoyance perception in scenarios involving multiple electric vehicle AVAS sounds. It changes the traditional isolated source-based view into a scene-based one by investigating the combined presence of multiple vehicle sounds as they are experienced in realistic traffic environments. Binaural listening tests were conducted using recorded electric vehicle pass-by sounds. The stimuli presented different traffic scenarios, including single and multiple-vehicles. Selected stimuli were spatially arranged to simulate vehicles approaching from opposite directions. After each stimulus, participants rated their perceived annoyance, enabling a comparison of annoyance responses between isolated and multiple AVAS sound scenarios. The test investigated how different levels of overlapping AVAS sounds affect perceived annoyance when multiple electric vehicles are passing by.

12. Empowering Long-form Omni-modal Understanding with Robust Audio Perception

Authors: Kaiying Yan, Luoyi Sun, Xiao Zhou, Weidi Xie

Categories: cs.LG Score: 5.60/10 (Obj:8 Id:5 Ind:5 Comp:5 Eff:6 Nov:5 Rep:8)

  • Strength: AVDC 数据集首次显式解耦 V/A/AV 三类 caption 并保留 37% 不可见声源事件，在 AVDC-test 上将 Qwen2.5-Omni 不可见声源 miss rate 从 90.7% 降至 77.5%，Video-SALMONN-2 total miss+hall 降至 37.6%（开源最优）。
  • Weakness: 绝对效果仍不可用（59.5% 整体 miss rate），audio QA 提升 <2pp 处于噪声范围，ablation 未隔离”解冻 encoder”与”AVDC 数据”的 confound，AVDC-test 存在”模型生成标注→模型评测”的循环风险，未与同期 Qwen3-Omni/Omni-Captioner 比较。
  • Full review: Claude Code 全文七公理审稿

Recent advances in large-scale multimodal models have drivenremarkable progress in vision-language tasks; however, comprehensiveomni-modal understanding remains under-explored, largely due to thescarcity of datasets with rich, explicitly aligned auditory cues. To bridgethis gap, we present AVDC (Audio-Visual Decoupled Captions), a large-scaledataset designed to disentangle visual and auditory semantics. Specifi-cally, we propose an automated pipeline that leverages off-the-shelf mod-els to annotate videos with tripartite captions: visual-only (V), audio-only (A), and joint audio-visual (AV). This decoupled structure explic-itly captures both modality-specific nuances and complex cross-modalinteractions. Building upon this, we introduce AVDC-QA-CoT, a Chain-of-Thought augmented question-answering dataset to foster audio-visualreasoning. To fully exploit these resources, we employ a two-stage train-ing paradigm: omni-modal caption generation pre-training on AVDC, fol-lowed by instruction tuning on AVDC-QA-CoT. Extensive experiments acrossdiverse downstream tasks, spanning video captioning, audio-centric anal-ysis, and omni-modal benchmarks, demonstrate consistent and signifi-cant performance gains, showing the efficacy of our proposed datasetsand training strategy in advancing omni-modal perception. Code anddataset are related on https://radiant0726.github.io/AVDC-web/.

13. Evaluating SSL and ViViT Architectures for Cross-Corpus Audio MOS Prediction via LODO Validation

Authors: Mustafa Ozan Duman, Ahmet Emir Dirik

Categories: eess.AS | Huggingface link: https://huggingface.co/mustafa-ozan-duman/wavlm-transformer-mos-english Github link: https://github.com/mustafa-ozan/audio_mos_prediction_SSL_ViViT_codes Score: 5.32/10 (Obj:7 Id:5 Ind:5 Comp:4 Eff:5 Nov:4 Rep:9)

  • Strength: 在 19 个数据集（130K+ 样本）上对 SSL 和 ViViT 架构进行了大规模系统性 LODO 基准测试，冻结 WavLM-Large + 2 层 Transformer 在 URGENT 2024 上实现了具有竞争力的 0.36 MSE，并发布了代码和权重。
  • Weakness: 没有新机制或新见解（冻结 > 微调的 OOD 泛化能力是已知的迁移学习原理）；SOTA 比较使用 ARECHO 估计值而非实际指标；缺失关键基线（WavLM+线性头, MOS-Bench）；ViViT 在多个 LODO 案例中显示出负相关；LODO 泛化差距非常大（未见数据 MSE 高达 2.53）。
  • Full review: Claude Code 全文七公理审稿

Automatic Mean Opinion Score (MOS) prediction is essential for evaluating large-scale synthetic speech and audio enhancement systems, yet models frequently struggle with domain shift. This study presents a comprehensive benchmarking of three architectural frameworks: Frozen Self-Supervised Learning (SSL-FRZ), Fine-Tuned SSL (SSL-FT), and a Video Vision Transformer (ViViT). Evaluation is conducted in two phases: Part I utilizes a consolidated corpus of 130,000 samples across 19 diverse datasets, while Part II focuses on a purified 17-dataset English-only corpus. To assess robustness, a systematic Leave-One-Dataset-Out (LODO) protocol is employed to quantify the generalization gap between seen and unseen distributions. Finally, the top-performing model is benchmarked against 18 state-of-the-art (SOTA) metrics using the ARECHO framework. Results demonstrate that an English-only purified corpus consistently yields higher predictive precision across all architectures. While SSL-FT achieves the highest performance on seen validation data, the SSL-FRZ model provides superior robustness on unseen distributions, achieving a competitive Mean Squared Error (MSE) of 0.36 on the URGENT 2024 benchmark-closely matching domain-optimized SOTA metrics (MSE 0.30). Although the ViViT architecture remains below SSL-based models in total capacity, it delivers stable results in English-only trials. LODO results confirm that while models perform significantly better on seen samples, frozen SSL embeddings combined with deep Transformer encoders offer the most stable and scalable solution for universal speech quality assessment. To support further research, the top-performing English-only SSL-Transformer model and weights are made publicly available via Hugging Face.

14. PC-Mix: Partial-Component Audio Spoofing Detection under Mixed Speech and Environmental Sound Conditions

Authors: Zhenshan Zhang, Xueping Zhang, Linxi Li, Yechen Wang, Ming Li

Categories: cs.SD, cs.CL Score: 5.26/10 (Obj:7 Id:5 Ind:4 Comp:4 Eff:5 Nov:5 Rep:8)

  • Strength: 首个包含 partially spoofed environmental sounds 的数据集（126,586 clips, 140.65h），通过 E0-E4 评测子集系统揭示 background-domain shift 导致 EER 从 0.67% 飙升至 9.15%，matched 条件下联合训练使 5-class accuracy 从 69.40% 提升至 85.12%。
  • Weakness: Baseline 覆盖严重不足——仅 MultiResoModel 一种 backbone，未对比 AASIST/CompSpoof/ESDD2 参赛方案，AudioLDM2 同时用于训练和 in-domain 评测，frame-level speech EER 仍高达 21.84%，方法为同团队 CompSpoof 框架的直接增量扩展。
  • Full review: Claude Code 全文七公理审稿

Recent studies on partial audio spoofing mainly focus on studio-recorded speech with temporal localization of spoofed segments. However, these studies often overlook realistic conditions where spoofed and bonafide segments simultaneously coexist across speech and environmental sound components. In this paper, we present PC-Mix, the first dataset for partial-component spoofing detection, where either or both audio components may be partially spoofed. In PC-Mix, bonafide and partially spoofed environmental-sound components are first constructed and mixed with speech signals from an existing partial-spoof dataset, producing audio in which either or both components may be locally manipulated. This design addresses two major gaps in existing partial spoofing benchmarks: the lack of realistic environmental sounds in speech partial spoofing scenarios and the absence of partial spoofing detection for environmental sound components. We further establish standardized evaluation protocols and design a joint learning framework to optimize spoofing detection across speech, environmental sound, and mixed audio. Experiments highlight the increased difficulty introduced by mixed conditions. The results demonstrate that training under matched target conditions is more effective than directly transferring models trained on speech or environmental sound components.

15. Efficiently Adapting Spoken Language Models for the Singaporean Context

Authors: Ng Jia Sheng Jason

Categories: cs.CL, cs.AI | 10 pages, 2 figures Score: 4.95/10 (Obj:6 Id:5 Ind:5 Comp:4 Eff:5 Nov:4 Rep:5)

  • Strength: 通过 LoRA + surrogate text-QA + CoBa 多任务训练，将 Voxtral-Mini-3B 在新加坡口音识别（24.7→72.6）和性别识别（15.7→93.9）上大幅提升，且仅损失约 1.3 个点的 Speech QA 能力，证明资源受限下 PEFT 适配 SLM 的可行性。
  • Weakness: 核心提升主要来自在 MNSC 大规模数据上的训练而非方法创新，Speech QA（论文核心 use case）实际从 76.4 降至 75.1，ASR 仍落后 MERaLiON-2，且 Spoken QA 评测集和训练数据均内部构建未公开，独立验证困难。
  • Full review: Claude Code 全文七公理审稿

Spoken language models (SLMs) unify speech perception and reasoning, but adapting them to sensitive domains is underexplored, especially when the original training data is inaccessible and the use case demands multilingual, spoken-query interaction. We adapt an open-source SLM to the Singaporean Home Team context across five speech tasks in Singapore’s four official languages, combining LoRA fine-tuning, a surrogate text-QA dataset that guards against catastrophic forgetting, and a multi-task objective that adapts the CoBa reweighting scheme to speech. We also build HTD-multilingual-QA, a 504,853 sample multilingual QA dataset in text and spoken form. The resulting HT-Moonstone (5B) matches or outperforms SLMs up to 7x its size on most tasks, attains the best accent and gender recognition among all models evaluated, and loses under 2\% of its original speech QA ability.

16. Which Languages Transfer Best to Warlpiri? A Similarity-Based Study for Low-Resource ASR

Authors: Pravina Mylvaganam, Eliathamby Ambikairajah, Ting Dang, Vidhyasaharan Sethu, Tuende Szalay

Categories: cs.CL, eess.AS | Accepted by Interspeech 2026 Score: 4.70/10 (Obj:5 Id:4 Ind:8 Comp:5 Eff:5 Nov:4 Rep:5)

  • Strength: 结合声学与语言学多维度相似性指导Warlpiri ASR迁移，Assamese作为源语言将WER从41.0%（多语言baseline）降至32.6%，并发现声学相似性主导fine-tuning、语音库相似性主导zero-shot的分化规律。
  • Weakness: 核心方法论（用embedding cosine similarity + linguistic features指导cross-lingual ASR transfer）已在Wu et al. (2021)、Kim et al. (2025)等工作中提出，本文仅限application-level novelty；11个语言的数据点统计效力不足，且未对比同期更强baseline（如Whisper-large-v3、omnilingual。
  • Full review: Claude Code 全文七公理审稿

This paper investigates how language similarity can improve cross-lingual transfer for automatic speech recognition (ASR) in extremely low-resource settings. Warlpiri, an Australian Aboriginal language, has very limited transcribed speech data, making transfer learning essential. We propose a framework combining acoustic similarity from pre-trained speech models with linguistic similarity based on typology, phoneme inventories, grammatical, and syntactic features to rank high-resource source languages and evaluate their effectiveness for ASR transfer to Warlpiri. Experiments with Whisper show that acoustically and typologically similar languages outperform monolingual and multilingual baselines. Assamese and Hindi achieve substantial reductions in word and character error rates. Correlation analysis further indicates that acoustic similarity is the strongest predictor of fine-tuning performance, while phoneme inventory and typological similarity better explain zero-shot transfer.

17. Graph Representation of RaagBase: A Unique Dataset for Hindustani Music

Authors: Chandan Misra, Swarup Chattopadhyay

Categories: cs.SD Score: 3.37/10 (Obj:5 Id:2 Ind:5 Comp:4 Eff:3 Nov:2 Rep:5)

  • Strength: 从权威文献（Bhatkhande）构建 notation-based raag 数据集，note frequency + similarity graph + Louvain/LPA 在最优 threshold 下达到 NMI 0.96/ARI 0.97。
  • Weakness: 仅 116 样本/3 raag，无任何 baseline 对比，方法为标准 similarity graph + 成熟聚类算法的直接应用，threshold 敏感且在低/高阈值下性能骤降，”novel graph representation”名不副实。
  • Full review: Claude Code 全文七公理审稿

Raag classification is a fundamental MIR task for Hindustani Music, with applications in recommendation, education, archiving, and intelligent search. However, raag clustering remains underexplored, as most existing approaches rely on annotated audio or labeled datasets. While annotated melodic phrases capture characteristic patterns, complete note sequences preserve temporal structure and contextual dependencies, making them more suitable for data-driven modeling. In this work, we introduce RaagBase, a notation-based text dataset consisting of note sequences from compositions by Pt. Bhatkhande. Furthermore we propose a novel graph-based representation of raag structures by modeling the dominance and absence of notes in compositions. Each composition is represented as a node, and the edges between two compositions corresponds the similarities between them based on the note frequency distribution. Further, we apply established graph clustering techniques to identify groups of similar raag compositions. Experimental results demonstrate highly coherent clusters with strong agreement to ground-truth raag labels, thereby validating both the dataset and the proposed representation. The dataset is publicly available at https://anonymous.4open.science/r/RaagBase-5427.