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。

2026-08-07
日期2026-08-07
已评分—
均分—
最高—

Daily Papers — 2026-08-07

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

1. How Much AI Is in This Track? Quantifying the Proportion of AI-Generated Stems in Hybrid Music Mixtures

Authors: Fernando Garcia de la Cruz, David López-Ayala, Pablo Zinemanas, Emilio Molina, Martín Rocamora

Categories: eess.AS, cs.AI, eess.SP | Accepted at International Society for Music Information Retrieval Conference (ISMIR 2026) Score: 6.60/10 (Obj:8 Id:6 Ind:5 Comp:8 Eff:6 Nov:8 Rep:8)

  • Strength: 首次将 AI 音乐检测从二分类重构为连续比例回归（α∈[0,1]），提出 codec-agnostic 的组合混合生成方法论，在同 pipeline held-out 测试中达到 MAE=0.076、R²=0.85，并发现乐器级检测灵敏度差异源于 codec artifact 的频率分布。
  • Weakness: 核心结论仅在单一 codec（EnCodec 3kbps）、同 pipeline 的 24 个 held-out tracks 上验证，无跨 codec/生成器/数据集的独立评测锚点，而二分类检测领域已证明跨平台泛化是核心挑战。
  • Full review: Claude Code 全文七公理审稿

AI-generated music is increasingly used at the stem level, with producers integrating synthetic drums, basslines, or vocals alongside human-performed instruments. However, current AI music detection systems are binary, treating tracks as either fully AI or fully human. In this paper, we reformulate AI music detection as a regression problem on a continuous AI energy ratio, alpha in [0, 1]. We propose a methodology that leverages a multi-track music dataset to assemble mixtures of human-performed and AI-reconstructed stems (obtained using a neural audio codec) with known proportions of each content type. Using this approach, we first show that a CNN-based model trained on fully AI-generated or human-performed tracks, which achieves >99% accuracy as a binary detector, when faced with mixed content, yields an output that rises with the AI stems’ energy contribution, acting as a noisy and miscalibrated estimator. Our analysis of the influence of different stems shows that detection sensitivity depends on the instrument and reflects its frequency content: drums and guitar carry strong codec-artifact signatures, while vocals and bass are less detectable. Based on these insights, we train a similar CNN-based model for regression of alpha, achieving MAE = 0.076 and R^2 = 0.85 on held-out mixtures from the same pipeline. These results suggest that the regression formulation is an initial promising step towards AI-music detection in realistic music production workflows.

2. SemBridge: Semantic Token Anchoring for Continuous-Latent Autoregressive Speech Generation

Authors: Hanke Xie, Haopeng Lin, Jiale Qian, Dake Guo, Yuepeng Jiang et al.

Categories: eess.AS, cs.SD Score: 6.30/10 (Obj:8 Id:8 Ind:7 Comp:8 Eff:6 Nov:6 Rep:8)

  • Strength: 在同一 0.8B backbone 下的控制实验证明，离散语义 token 分类目标显著优于连续特征回归（CER 1.47 vs 1.90），且训练时锚定、推理时移除的设计不增加推理开销，消融中 anchoring 将 ZH-Hard CER 从 16.87 降至 11.87（29.6% 相对降低）。
  • Weakness: 系统级对比中 speaker similarity 系统性低于 VoxCPM/VoxCPM2（ZH 0.758 vs 0.772/0.795），ZH-Hard CER 不及 VoxCPM2 (9.79 vs 8.13)，且 Table 1 vs Table 2 中 SA-VAE 对 ZH CER 的影响方向不一致（Table 1 变差 1.52→1.63，Table 2 变好 1.58→1.51），未充分解释。
  • Full review: Claude Code 全文七公理审稿

Continuous-latent autoregressive speech generation has emerged as a promising alternative to discrete-token modeling by avoiding quantization loss and preserving richer acoustic information. However, continuous acoustic targets do not ex- pose linguistic structure as explicit token-level prediction tar- gets. Consequently, the autoregressive language model (LM) must acquire linguistic structure indirectly through acous- tic prediction, which can compromise the content fidelity of generated speech. We propose SemBridge, a training-only semantic-token anchoring framework for continuous-latent autoregressive speech generation. SemBridge uses discrete se- mantic tokens to directly supervise autoregressive LM states and employs a Semantic-Aligned Acoustic VAE to organize the continuous target space under the same semantic refer- ence. The semantic supervision is used only during train- ing, while inference remains entirely continuous. We evalu- ate SemBridge on zero-shot text-to-speech (TTS) and score- conditioned singing voice synthesis (SVS). Across multi- ple benchmarks, SemBridge improves content accuracy, as measured by word and character error rates (WER/CER), while maintaining competitive speaker similarity and percep- tual quality. Experimental results demonstrate that explicit semantic-token supervision for autoregressive state learning is an effective and general direction for continuous speech generation. Speech samples are available.1 The model code and checkpoints will be available at https://github.com/ASLP- lab/SemBridge

3. Do Audio Language Models Use Paralinguistic Evidence? Counterfactual Audits for Response Evaluation

Authors: Kevin Miller, Arjun Chandra, Venkatesh Saligrama

Categories: cs.CL Score: 6.30/10 (Obj:8 Id:8 Ind:6 Comp:6 Eff:6 Nov:5 Rep:6)

  • Strength: 提出干净的反事实审计框架（transcript-fixed counterfactual），通过 native vs. contrastive 协议对比和 perception/response-mapping 分解，在 Gemini/GPT/开源模型上发现 contrastive success 系统性高估 native judge 可靠性且相似精度掩盖不同失败模式。
  • Weakness: 审计项为 TTS 合成（外部效度未验证），核心问题和审计方法与 VoxParadox (2026-05)、Auditing Protocol-Level Shortcuts (2026-07)、ParaPairAudioBench (2026-06) 等近期/并发工作高度重叠，主要结果依赖闭源 API 模型影响长期可复现性。
  • Full review: Claude Code 全文七公理审稿

Audio-language models (ALMs) are increasingly used as judges for speech-to-speech systems, but a judge that receives audio may not actually use paralinguistic evidence. We introduce counterfactual audits for paralinguistic response evaluation. Each audit item holds the transcript fixed while varying affect, prosody, or the timing of an affective shift, forcing a valid judge to track the audio cue rather than lexical content or response style. We evaluate ALM judges using a native one-context judgment protocol and a contrastive recoverability control, then further decompose each item into its constituent perception and response-mapping skills. This yields useful diagnostic states that identify different sources of judge failures. Across Gemini, GPT, and open audio models, we find that contrastive success often overstates native judge reliability, and that similar aggregate accuracies can hide different failure modes. These results suggest that ALM judges should not be evaluated by accuracy alone, instead requiring thorough behavioral audits before deployment.

4. MMAG: A Multi-Control Mixed Audio Generation Benchmark

Authors: Zihao Zheng, Xuenan Xu, Jiahao Mei, Yixuan Li, Minghao Lv et al.

Categories: cs.SD, eess.AS | 15 pages, 6 figures Score: 6.20/10 (Obj:8 Id:6 Ind:5 Comp:6 Eff:8 Nov:6 Rep:6)

  • Strength: MMAG 首次将 speech+music+sound effects 混合音频生成的评测统一到 3 个子集（4k main + 0.7k voice cloning + 1.8k timestamp）和 4 个维度（acoustic fidelity、speech quality、semantic consistency、temporal control），揭示了当前 10 个代表性模型均无法同时兼顾所有维度（如 Dasheng-Au。
  • Weakness: 评测的独立性存在结构性缺陷：MMAG 作者与被评测系统 AuDirector 共享 3 位作者且为其做了专门 timestamp adaptation；标注和 WER 评测共享 Whisper 模型家族；AAJ 指标依赖 LLM judge 但未验证其独立性。遗漏 UniSonate (ACL 2026) 等同期混合音频生成 baseline。
  • Full review: Claude Code 全文七公理审稿

Recent audio generation systems have progressed from single-modality synthesis to generating complex acoustic scenes containing speech, music, and sound effects. Therefore, evaluating these models requires assessing multiple interacting capabilities, including semantic fidelity, speaker consistency, and temporal control, yet existing benchmarks focus on isolated domains or coarse-grained descriptions. To address this gap, we introduce the Multi-control Mixed Audio Generation (MMAG) benchmark. MMAG contains approximately 4,000 manually verified audio clips with rich annotations covering speech content, speaker identity, music attributes, sound events, and temporal relationships, together with dedicated subsets for voice cloning and timestamp-conditioned generation. We further propose a systematic evaluation protocol that measures acoustic fidelity, speech quality, semantic alignment, and temporal accuracy. Benchmarking representative agentic orchestrators, unified audio-visual generation models, and native mixed-audio generators reveals substantial performance trade-offs across these capabilities, with no existing model performing consistently well. Our results highlight the remaining challenges of controllable mixed audio generation and establish MMAG as a comprehensive benchmark for future research.

5. Comparative Analysis of Spread-Spectrum Codes for Fibre-Optic Distributed Acoustic Sensing

Authors: Maria Freire-Hermelo, Elie Awwad

Categories: eess.SP, physics.optics Score: 6.00/10 (Obj:8 Id:6 Ind:8 Comp:5 Eff:6 Nov:5 Rep:5)

  • Strength: 首次在统一仿真框架下系统比较 5 类编码序列（含 ZC 的 3 个频率多样性等级）在 DAS 中的性能，发现频率多样性是相位噪声鲁棒性的关键因素（高多样性序列在 10 Hz 线宽/1.5 km 下相位误差 2.7×10⁻³ rad vs 低多样性 5.3×10⁻³ rad）。
  • Weakness: 纯仿真无实验验证，”comprehensive” 声称被缺失 2025 年新编码（Polyphase, Genuine Two-Phase Perfect Code, P4-code）削弱，频率多样性缺乏定量定义和理论解释，无代码开源。
  • Full review: Claude Code 全文七公理审稿

This paper presents a comprehensive performance comparison of several coded sequences for distributed acoustic sensing systems through numerical simulations. Correlation analysis reveals that perfect autocorrelation sequences provide accurate channel estimation. Phase noise sensitivity varies significantly with frequency diversity, with high-diversity sequences maintaining lower estimation errors for phase and intensity despite increased laser linewidths. For a coded distributed acoustic sensing system with \qty{10}{\hertz} laser linewidth, high-diversity sequences achieve $2.7\times10^{-3}$ rad phase error compared to $5.3\times10^{-3}$ rad for low-diversity codes for a \qty{1.5}{\kilo\meter} fibre length. The findings establish performance trade-offs between correlation quality, phase noise resilience, and practical implementation constraints for next-generation fibre-optic distributed acoustic sensing systems.

6. Cloud-Boosted Low-Compute Multi-Channel Speech Enhancement

Authors: Xulin Fan, Juan Azcarreta, Ashutosh Pandey, Jesus Alvarez, Ke Tan et al.

Categories: cs.SD, cs.LG | Accepted to Interspeech 2026 Score: 5.70/10 (Obj:8 Id:6 Ind:8 Comp:5 Eff:6 Nov:5 Rep:5)

  • Strength: 在 64ms 通信延迟下实现边缘端 SI-SDR 提升 3.77 dB（Standard）/ 3.49 dB（Challenging），仅增加 1.5% 参数和 2.4% MACs，且显著优于参数量翻倍的 TinyGRU-Large。
  • Weakness: PESQ 几乎无提升（2.29→2.33），未与原始 Knowledge Boosting 方法在相同设置下直接对比，未开源代码，仅在模拟数据上评测，三组件中 delayed input concatenation 不构成真正 novelty。
  • Full review: Claude Code 全文七公理审稿

Low-latency, low-compute speech enhancement is essential for wearable devices with real-time communication requirements, but strict computational constraints significantly limit on-device performance. Knowledge Boosting has been proposed as an effective approach to improve edge model performance by leveraging a more capable server-side model, but performance gains for speech enhancement have been limited. We propose a collaborative framework incorporating three techniques: (1) delayed server output as additional input, (2) layerwise feature boosting that transfers intermediate server representations to guide edge inference, and (3) collaborative multichannel Wiener filtering, which fuses weighted covariance matrices estimated from both server and edge models for improved beamforming. Experimental results demonstrate that the proposed collaborative framework significantly outperforms the edge-only baseline with minimal additional computational overhead.

7. AVCap: Reinforcing Audio-Video Joint Caption with Detail-Aware Reward

Authors: Mingyang Wu, Kaituo Feng, Bohao Li, Kaixiong Gong, Zihao Yin et al.

Categories: cs.CV Score: 5.70/10 (Obj:8 Id:6 Ind:2 Comp:5 Eff:6 Nov:5 Rep:8)

  • Strength: 提出原子级 QA-verification reward (Da-GRPO) 用于音视频联合字幕训练，AVCap-30B 在外部 benchmark UGC-VideoCap 上达到 85.1 分，在 WorldSense 上 34.3 分超越 Gemini-2.5-Pro (33.8)，数据/方法/基准三者形成完整框架。
  • Weakness: 训练 reward 和主评测指标 AVCap-Score 使用同一 judge model (Qwen3-30B-A3B) 执行相同的 raise-answer-check 流程，构成 fatal 级别循环论证；30B 规模缺少同规模开源 baseline 的公平对比。
  • Full review: Claude Code 全文七公理审稿

Detailed audio-video joint captioning is essential for multimodal video understanding and generation. However, prior works are constrained by three main limitations: (1) the scarcity of high-quality public datasets with fine-grained audio-visual joint captions; (2) reinforcement-learning methods that rely on coarse reward signals; and (3) the lack of a benchmark and metric for evaluating detailed audiovisual captions at the atomic level. To address these challenges, we propose: (1) AVCap-100K, a high-quality dataset of 100K temporally aligned, detail-rich audio-video captions; (2) AVCap, a model optimized via Detail-Aware GRPO (Da-GRPO) that achieves state-of-the-art performance among open-source models and matches or surpasses proprietary models on several evaluations; and (3) AVCap-Bench and AVCap-Score, a specialized benchmark and metric for evaluating atomic-level details in audiovisual captions. Our code, models, and datasets are available at https://huggingface.co/collections/Apryle/avcap.

8. Assessing AI-generated music detection in real-world broadcast monitoring

Authors: David López-Ayala, Fernando García de la Cruz, Pablo Zinemanas, Emilio Molina, Martín Rocamora

Categories: eess.AS, cs.AI | Accepted for ISMIR 2026 Score: 5.21/10 (Obj:8 Id:5 Ind:4 Comp:5 Eff:5 Nov:5 Rep:5)

  • Strength: 首个真实广播 AI 音乐检测数据集 BAMM（40h），清晰暴露合成→真实广播的 domain gap，CNN Broadcast 在 RTB 上 ROC AUC=0.775 但 F1 仅 0.472，证明当前方法不可靠。
  • Weakness: 标注循环依赖（CNN Clean 同时参与标注和评估），仅评估 1 种架构未覆盖 SpecTTTra 等 SOTA baseline，绝对性能远低于可用水平且标签 pipeline 依赖私有组件。
  • Full review: Claude Code 全文七公理审稿

The proliferation of AI-generated music in broadcast media raises concerns about transparency and fair compensation, but reliable detection under real broadcast conditions remains unresolved. Existing studies report substantial performance degradation in this domain, yet their evaluations are limited to synthetic broadcast data. To address this gap, we introduce BAMM (Broadcast AI-Music Monitoring), a 40-hour dataset of real-world television recordings containing AI-generated and human-made music. We compare clean-trained and broadcast-trained CNN variants across three progressively more challenging scenarios: Clean Foreground Music (CFM), Synthetic TV Broadcast (STB), and Real TV Broadcast (RTB). Both models achieve near-perfect performance on CFM but degrade substantially under synthetic broadcast conditions. Broadcast-oriented training improves robustness compared with clean training, although performance remains limited. On RTB, evaluated using BAMM, both models degrade further and show substantial score overlap between AI-generated and human-made music. These results expose a critical domain gap and show that current training approaches on CNN-based detectors remain insufficient for reliable AI-generated music detection in broadcast monitoring.

9. Beyond Call and Response: Modelling Reciprocal Coordination in Human-AI Vocal Ensembles

Authors: Polina Proutskova

Categories: cs.HC, cs.SD | 5 pages, 2 figures Score: 3.16/10 (Obj:5 Id:2 Ind:5 Comp:4 Eff:2 Nov:4 Rep:1)

  • Strength: 论文从民族音乐学视角正确定位了无指挥声乐合唱中”互惠协调”问题的独特性，将非等节拍视为通用框架的硬案例具有理论洞察力，VocalLanes 的手机多通道采集方法具有田野实用性。
  • Weakness: 论文遗漏了多篇直接相关的前置工作（Hyodo et al. 2024 歌手交互建模、Frontiers 2024/2026 人类-AI 合唱协调动力学实证研究），且全部核心模块（状态推断、agent 策略、评测）均处于”proposed”阶段，无可评估的实验结果。
  • Full review: Claude Code 全文七公理审稿

Musical interaction with AI is often organised as a response loop: a human performs, the system interprets that action, and the system answers, accompanies, or schedules a musical event. Unconducted vocal ensembles pose a different problem. Singers act simultaneously and continuously affect one another; neither timing nor pitch is fixed by a conductor, metronome, accompaniment, score, or tuning source. Collective organisation emerges from many-to-many reciprocal adjustment. This paper frames such ensembles as coupled dynamic systems and proposes a research architecture for vocal agents that enter, rather than merely track, their collective states. Some target repertoires are metrical, while others exhibit non-isochronous temporal contours that cannot be reduced to a beat grid; we treat the latter as a hard case for a general framework. The architecture connects multichannel capture in the field to dialect- and singing-aware representation, collective-state inference, vocal generation, and in-situ evaluation. The resulting agenda asks not only whether an artificial singer can synchronise, but how its presence reorganises human coordination, leadership, style, and musical transmission.