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。

2026-07-28
日期2026-07-28
已评分—
均分—
最高—

Daily Papers — 2026-07-28

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

1. faster-enhancer.c: A Dependency-Free int8 Runtime for Streaming Speech Enhancement on Commodity CPUs

Authors: Gyeongmin Kim

Categories: eess.AS, cs.SD | 5 pages, 2 figures, 2 tables. Code: https://github.com/kdrkdrkdr/faster-enhancer.c Score: 6.95/10 (Obj:9 Id:6 Ind:9 Comp:8 Eff:8 Nov:6 Rep:9)

  • Strength: 围绕单一固定模型专门化 int8 运行时，M2 单核 0.069 RTF（3.3× vs fp32 ONNX Runtime），质量损失仅 -0.006 PESQ / -0.08 dB SNR，跨 SIMD 层级位一致，代码零依赖开源。
  • Weakness: 基线仅 fp32 ONNX Runtime，未对比通用 int8 推理库（XNNPACK/KleidiAI）；部分融合缺乏开关消融；手机端 30 分钟长跑越过预算（1.77% 死线错过），部署可用性受限。
  • Full review: Claude Code 全文七公理审稿

This is an implementation and measurement study of what it costs to run a streaming speech enhancer on a CPU. We port FastEnhancer-Medium at 48 kHz to faster-enhancer.c, a C runtime with six int8 GEMM tiers selected at initialization, leaving architecture and weights untouched. One Apple M2 core reaches 0.069 real-time factor, against 0.230 for the fp32 ONNX Runtime graph on the same machine, a 3.3x speedup. A Galaxy S23+ (Snapdragon 8 Gen 2) reaches 0.096. The speedup comes from specializing every layer of the runtime around one fixed model. Activation ranges are recomputed per frame, so no calibration set is needed; the k=3 convolutions use Winograd F(2,3); cross-stage state is fp16; the GRU and the dequantization epilogues are fused; and nothing is allocated after startup. Over 824 VoiceBank-DEMAND utterances the engine tracks fp32 to within -0.006 PESQ and -0.08 dB SNR. Speed alone does not settle deployment cost. The enhancer holds a fraction of a core for as long as the microphone is open, so its real-time factor is a duty cycle. A benchmark races through a file; an audio callback does not. Pacing to the 6.67 ms deadline costs 4.2x per frame, saves 49% of the energy, and leaves the cheapest core placement missing 96% of its deadlines. All SIMD tiers within an architecture family emit byte-identical output. The runtime is released as a dependency-free library.

2. From Semantics to Readout: Mechanistic Understanding of Audio Tokens after Fine-Tuning for Temporal Audio Grounding

Authors: Yujian Ma, Jinqiu Sang, Ruizhe Li, Jiaao Yu, Ang Li

Categories: cs.SD Score: 6.84/10 (Obj:8 Id:9 Ind:6 Comp:8 Eff:6 Nov:8 Rep:5)

  • Strength: 通过 cross-checkpoint control（decoder-swap vs state-swap）和 size-matched residual-delta erasure 等严格因果识别，证明 grounding fine-tuning 主要改善 decoder readout 而非创建 event evidence，base checkpoint 已含可恢复的 latent event structure。
  • Weakness: 仅在 2 个同家族 Qwen 模型、1 个 fine-tuning 任务上验证，标题声称的 “Mechanistic Understanding of Audio Tokens after Fine-Tuning” 外延远大于验证范围；代码未确认开源，关键量化结论仅基于 100 实例。
  • Full review: Claude Code 全文七公理审稿

Large audio-language models (LALMs) convey acoustic evidence to language decoders through native audio tokens, yet the internal roles of these tokens remain poorly understood. Using temporal audio grounding as a diagnostic setting, we examine how language-model fine-tuning affects the layerwise semantics, decoder accessibility, and temporal output alignment of native audio-token states through four complementary analyses: query-conditioned token semantics, calibrated token readout, temporal-window probes, and residual-delta erasure during generation. Alongside substantial improvements in temporal localization, semantic analysis of Qwen2.5-Omni shows that latent evidence for queried events is already present before fine-tuning and that the audio tokens most strongly aligned with the queried event appear at similar temporal positions before and after fine-tuning. After fine-tuning, event-related information in audio tokens becomes more accessible to the decoder, especially in early and middle layers, and a cross-checkpoint control shows that this improvement arises primarily from decoder adaptation. Temporal probes show that the base checkpoint already contains recoverable information about annotated windows and that fine-tuning mainly improves alignment with each checkpoint’s own predicted temporal support. Residual-delta erasure further shows that removing audio-token updates within predicted windows harms timestamp generation more than removing the same number of randomly selected updates. The same broad improvements in decoder readability and prediction alignment also appear in Qwen2-Audio. Together, these results support a semantics-to-readout account in which grounding fine-tuning helps the decoder read existing event evidence and connect it more reliably to temporal outputs.

3. GraphIDyOM: A graph-native Python reimplementation of IDyOM for musical expectation modelling

Authors: Lluc Bono Rosselló

Categories: cs.SD, q-bio.NC | 18 pages, 7 figures Score: 6.80/10 (Obj:6 Id:9 Ind:9 Comp:5 Eff:8 Nov:5 Rep:9)

  • Strength: 在 12 种 viewpoint 配置、9227 events 上对 Lisp IDyOM 达到 ΔIC ≤ 0.003 bits、r ≥ 0.9997 的比特级忠实度，且预测延迟 0.8–1.7 ms/event（比 IDyOMpy 快 ~50×），同时首次将 LTM/STM 内存暴露为可导出 GraphML 的显式图对象。
  • Weakness: 所有验证仅限于单一 Bach chorale 语料，三个扩展（网络分析、PPM-decay、交互式 server）均为已有概念的工程集成且仅以单篇示例演示，无机制创新或可迁移经验规律，新颖性增量薄。
  • Full review: Claude Code 全文七公理审稿

The Information Dynamics of Music model (IDyOM) has played a central role in computational accounts of musical expectation by providing event-by-event estimates of uncertainty and surprise from symbolic musical sequences. However, its reference implementation is difficult to integrate with contemporary Python workflows, and its internal memory structures are not easily accessible for inspection or modification. We introduce GraphIDyOM, a graph-native Python reimplementation of IDyOM that represents long-term and short-term predictive memories as explicit graph objects while preserving the model’s variable-order, multiple-viewpoint architecture. GraphIDyOM returns event-wise information content and entropy, exposes internal memory structures for analysis and export, and supports access through a local server. We validate the implementation against the original Lisp IDyOM across single, projected, and multiple-viewpoint configurations, and benchmark its coverage and computational performance against a recent reimplementation. We then demonstrate how the explicit memory representation supports network analysis of learned memories, projection of expectation values onto musical networks, recency-sensitive memory retrieval, and interactive applications. GraphIDyOM therefore provides both a faithful and accessible reimplementation of a widely used model and a platform for studying musical expectation through memory, topology, and interaction.

4. Less is More: Modality-Decoupling for General AIGC Audio-Video Detection

Authors: Jielun Peng, Yabin Wang, Yaqi Li, Jincheng Liu, Xiaopeng Hong et al.

Categories: cs.CV, cs.AI | First place in the General AIGC Audio-Video Detection Challenge at the IJCAI-ECAI 2026 DDL 2.0 Workshop Score: 6.42/10 (Obj:8 Id:5 Ind:6 Comp:5 Eff:8 Nov:5 Rep:8)

  • Strength: 在 DDL-GAV 竞赛获第 1 名（score 0.8460），FakeAVCeleb 达 ACC 0.998/AUC 0.999 SOTA；Fig.1 量化证明跨模态对应假设在一般场景失效（AUC 40.28%），为决策级融合提供了有力的经验动机。
  • Weakness: 缺少决策级 vs 特征级融合的 head-to-head 直接对比（相同 backbone），竞赛前 5 名差距仅 0.0022 无统计检验，且核心发现（跨模态对应在一般场景失效）已被 MVAD 数据集论文 (arXiv:2512.00336) 先前讨论。
  • Full review: Claude Code 全文七公理审稿

Generative AI has rapidly expanded audio-visual forgery beyond human-centric deepfakes into general scenes. Existing AIGC detection methods assume audio-visual content correspondence, identifying forgeries by spotting cross-modal inconsistencies. However, we empirically find that this assumption does not consistently hold in general scenarios. We argue that, for general audio-visual AIGC detection, decision-level fusion is a more robust alternative to feature-level fusion. Therefore, we propose DAV-Det, a decoupled audio-visual AIGC detection system that independently models forensic evidence from each modality. The visual detector leverages multi-granularity representations at global, patch, and segment levels to capture spatial forgery cues, while the audio detector exploits both temporal and spectral irregularities via a gated temporal-spectral dual-branch architecture to model acoustic artifacts. Our method ranks 1st in the General AIGC Audio-Video Detection Challenge of the IJCAI-ECAI 2026 DDL 2.0 Workshop, with a final score of 0.8460. Code is available at https://github.com/tuffy-studio/DAV-Det.

5. Finding the noise: Zero-shot AI Music Detection

Authors: Darius Afchar, Romain Hennequin

Categories: cs.SD | preprint – may be modified for a future publication Score: 6.37/10 (Obj:8 Id:5 Ind:6 Comp:8 Eff:5 Nov:5 Rep:5)

  • Strength: 在真实紧迫的 AI 音乐检测问题上，首次将 fakeprint 表示扩展到零样本设置，用 NMF 字典学习+Gaussian blur 重建误差实现一类检测，在 3 个数据集 18 个 AI 模型上多数类达到 95-100% 准确率，且方法无需深度学习、计算快速。
  • Weakness: 无任何 baseline 比较（包括最简异常检测 baseline），2/18 类完全失败（Mubert 5%, Mureka 14.9%），高精度场景（1% FPR）下多个类崩溃，核心方法由已有组件直接组合且 concurrent work MusicDET (ICML 2026) 已用更 principled 的 normalizing flows 做同类零样本任务，代码和数据暂未公开。
  • Full review: Claude Code 全文七公理审稿

We present a novel method for AI-generated music detection in scenarios where the models that generated the input samples are unknown to the detector (e.g., from a newly released service). Since 2023, there has been a multiplication of user-friendly AI-music generation services (e.g., Suno, Udio), along with regular updates and new features. There is thus a need to address synthetic content detection in an unsupervised way to adapt to this rapidly changing context. This angle has not been much studied in music yet. We propose to study two tasks. First, discriminating between real and synthetic music. This may be approached in a one-class manner, namely, using some baseline real music and trying to determine what falls outside. Second, zero-shot multi-class identification, which is more similar to an unsupervised clustering task on a mix of real and various AI-music generations, where the goal is to create coherent, high-purity clusters. We propose a combination of a previously proposed artifact-extraction method, on top of which we apply non-negative matrix factorization and simple classification and clustering methods. We achieve excellent performance on both tasks, showing that the proposed methods may be used to monitor large-scale catalogs that may receive AI-generated samples from various newly released generative models.

6. Extracting Voice Styles from Frozen TTS Models via Gradient-Based Inverse Optimization

Authors: Gyeongmin Kim

Categories: eess.AS, cs.SD | 5 pages, 2 figures, 2 tables. Code: https://github.com/kdrkdrkdr/supertonic.embed Score: 6.30/10 (Obj:8 Id:6 Ind:8 Comp:8 Eff:6 Nov:5 Rep:8)

  • Strength: 在 154 个说话人上，ECAPA-TDNN 相似度从 0.132 提升至 0.413（所有说话人一致改善，Wilcoxon p < 10⁻²⁶），验证器 EER 点接受率从 1.3% 升至 52.6%，方法仅需一段裸录音、无需转录本或对齐。
  • Weakness: 仅在单一 TTS 模型（SupertonicTTS 2）上验证，未与最接近的同类工作 Marras et al. [13] 做实验对比，核心技术由 VoiceLoop 的冻结适配 + Aldeneh 的 SSL 统计量描述符 + GAN inversion 范式组合而成，机制层面增量有限。
  • Full review: Claude Code 全文七公理审稿

Some text-to-speech systems ship a synthesis model and preset style vectors but not the reference encoder that turns audio into such a vector. The model still accepts a style vector; a user with a voice of their own cannot produce one. We solve for that input directly, inverting the released pipeline by gradient descent: every weight stays frozen and only the style vector is optimized, against time-pooled WavLM statistics of one recording. Because the objective discards the time axis, the synthesized text may differ from the recording, so no transcript and no alignment are needed. On 154 speakers from two corpora, ECAPA-TDNN similarity rises from 0.132 to 0.413 and ResNet from 0.099 to 0.401, improving for every speaker; a verifier at its equal-error point accepts 53% of the recovered voices as the target, against 1% for the presets they start from.

7. Joint Text-Audio Alignment for EEG-to-Text Decoding in Chinese Speech Production and Perception

Authors: Tian Zheng, Xurong Xie, Xinxin Zhu, Xiaolan Peng, Feng Tian

Categories: cs.AI Score: 6.20/10 (Obj:8 Id:6 Ind:8 Comp:5 Eff:8 Nov:5 Rep:6)

  • Strength: 双轴对齐（BGE-M3 text + wav2vec 2.0 audio + CTC）在中文 EEG-to-text closed-set 分类上实现 SOTA：RA 82.37%、PL 41.43% Top-1（101 candidates），大幅超越所有 baseline，且消融发现单轴对齐反低于 CTC-only、双轴才互补有效。
  • Weakness: 核心方法为已有组件组合（CTC + InfoNCE + adapter + curriculum），无新模块或机制解释；未引用同年 2603.12628 的同类中文 production+perception 解码工作；代码未开源，curriculum 具体调度参数未完整报告。
  • Full review: Claude Code 全文七公理审稿

Decoding speech information directly from scalp electroencephalography (EEG) into text provides a potential non-invasive neural communication pathway for individuals with severe speech and motor impairments. Compared with invasive approaches such as electrocorticography, EEG is safer and more widely deployable, yet substantially more challenging to decode.This challenge is exacerbated for Chinese sentence decoding, which must handle a high-dimensional output space with thousands of characters, severe inter-subject variability, and low signal-to-noise ratios for text alignment.Existing methods commit to a single supervisory axis—either text semantics or audio acoustic features—yet neither can simultaneously satisfy the demands of sentence-level discriminability and fine-grained temporal resolution required for large-vocabulary Chinese decoding. We introduce EEGAlign, a novel parameter-efficient framework that jointly aligns EEG with two axes—text alignment with BGE-M3 text embeddings and audio alignment with wav2vec~2.0 speech features via contrastive learning followed by CTC character-sequence decoding. On ChineseEEG-2 data, EEGAlign yields state-of-the-art closed-set sentence classification performance, reaching up to 82.37% Top-1 accuracy on Reading Aloud EEG and 41.43% on Passive Listening EEG out of 101 candidates. Ablation studies show that the two alignment axes are highly complementary: combining them yields consistently better performance than either alone. To the best of our knowledge, this is the first study on decoding large-vocabulary Chinese sentences from non-invasive EEG during overt speech production, and achieving strong classification performance with relatively large closed-set candidate-sentence setting.

8. VAD to the Bone: Ultra-Tiny Speech Activity Detection for Edge Deployment

Authors: Stephen Bauer, Sheila Seidel, Shanza Iftikhar, Scott Veidenheimer, Gorkem Ulkar

Categories: eess.AS, cs.LG | Accepted for publication at INTERSPEECH 2026 Score: 5.84/10 (Obj:8 Id:5 Ind:8 Comp:6 Eff:5 Nov:5 Rep:6)

  • Strength: 在严格因果 200ms 条件下用 2.1k 参数 CNN-only 模型达到 0.850 AUC（AVA-Speech），INT8 量化无损，angle-aware QAT 在 INT4 上比标准 QAT 提升 1–4%。
  • Weakness: “SOTA” claim 依赖自定义四条部署要求（R1–R4）排除 AtomicVAD（因果 AUC 0.869 > kiloVAD 0.862/0.850），缺少最简 baseline，核心新颖性限于 angle-QAT 增量改进且绝对增益微调级。
  • Full review: Claude Code 全文七公理审稿

Voice activity detection (VAD) triggers downstream speech processing in always-on systems under strict memory, latency, and compute constraints. Recent compact models report strong accuracy but rely on components that are not widely supported: learnable filterbanks, recurrent layers, or non-causal post-processing. We propose kiloVAD, designed for embedded inference using standard Mel features, CNN-only layers, and tunable context/spectral parameters. We introduce per-layer structured pruning with self-distillation and angle-based quantization-aware training (QAT) that outperforms standard QAT by 1-4%. Evaluated per-frame under causal conditions, kiloVAD achieves 0.850 AUC on AVA-Speech with 2.1 k parameters and 200 ms context, establishing a new state of the art for causal, deployment-ready VAD.

9. Spacing Out: On the Reliability of Binaural Music Source Separation Metrics

Authors: Richa Namballa, Magdalena Fuentes

Categories: eess.AS, cs.SD, eess.SP | 6 pages + references, 6 figures, 1 table, 27th International Society for Music Information Retrieval (ISMIR) Conference Score: 5.79/10 (Obj:8 Id:6 Ind:8 Comp:5 Eff:5 Nov:5 Rep:8)

  • Strength: 通过 26 人感知实验首次验证 binaural MSS 空间指标与人类判断的一致性，发现 ΔILD 和 SRR 比 ΔITD 更可靠，并定量揭示 GCC-PHAT 在 bass（窄带）上交叉相关峰塌缩至 0 lag 的失效模式。
  • Weakness: 核心分析工具（PHAT-β, masked GCC-PHAT）均为已有方法迁移，未提出替代指标或解决方案；感知实验规模偏小（96 comparisons），且与前序工作（同一团队 ISMIR 2025）高度连续，独立增量有限。
  • Full review: Claude Code 全文七公理审稿

Despite the rising popularity of immersive audio, binaural music remains underexplored in music information retrieval (MIR), particularly regarding the task of music source separation (MSS). While existing stereo MSS models can process binaural audio, they often degrade the spatial quality of the separated stems and undermine listener immersion. Through a perceptual study comparing binaural and stereo MSS outputs, we evaluate how well objective spatial distortion metrics correlate with human perception. Our findings reveal varied agreement between these metrics and human judgment, highlighting a lack of reliability when used to evaluate binaural music tasks. Specifically, we find that Interaural Time Difference (ITD) estimation is highly sensitive to noise and separation artifacts. In evaluating two alternative ITD estimation methods, we uncover a critical trade-off between robustness and accuracy, particularly for narrow-band instruments like bass. These results underscore the need for accurate, interpretable spatial metrics designed for binaural music to develop models that preserve source localization and listener immersion.

10. Depression Markers in Speech: An Approach based on Tract Variables Dynamics

Authors: Sahar Altalhi, Tanaya Guha, Alessandro Vinciarelli

Categories: eess.AS, cs.AI | Accepted for publication in the Journal of the Acoustical Society of America (JASA) Score: 5.70/10 (Obj:8 Id:5 Ind:8 Comp:6 Eff:5 Nov:5 Rep:8)

  • Strength: 首次将三个非线性动力学度量（LLE、CD、SE）应用于声道变量序列检测抑郁，在临床确诊数据集（Androids Corpus, 64抑郁+54对照）上Reading Task分类达73.2%准确率，Cliff’s delta最高0.51。
  • Weakness: 缺乏与已有TV-based抑郁检测方法（Seneviratne 2020/2021）的直接对比，Interview Task分类性能（68.2%）低于LLDs baseline（73.3%），且部分统计显著性受样本量驱动而非真实效应。
  • Full review: Claude Code 全文七公理审稿

This study identifies new depression biomarkers based on the dynamical properties of tract variables, which represent geometric features describing the configuration of the speech articulators. A key advantage of this approach lies in its ability to quantify aspects of the articulatory process that have not been previously explored in the context of depression, namely predictability, complexity, and randomness. These properties are respectively characterised using the Largest Lyapunov Exponent, the Correlation Dimension, and the Sample Entropy. Thorough experiments were conducted on the Androids Corpus, a publicly available dataset comprising 64 speakers diagnosed with depression by clinicians and 54 control speakers with no reported history of mental health conditions. The results indicate that the proposed biomarkers effectively discriminate between the depressed and control speakers, as evidenced by the high Cliffs delta values across both read and spontaneous speech.

11. AMRD: Adaptive Multi-Teacher Relational Distillation for Lightweight Speech Emotion Recognition

Authors: Yuqi Li, Yi-Cheng Lin, Xianglong Wang, Kuo Yang, Xiaoqin Feng et al.

Categories: cs.LG Score: 5.60/10 (Obj:8 Id:5 Ind:9 Comp:5 Eff:5 Nov:4 Rep:6)

  • Strength: 在 7/8 学生-数据集设置上，SVM 自适应加权 + RSMD 关系蒸馏的组合优于最佳单教师 baseline（提升达 2.9% UA），且两组件在消融中均显示独立贡献。
  • Weakness: 两个核心组件均基于已有技术（one-class SVM 评分、SPKD-style 相似度矩阵匹配），新颖性增量有限；缺少 entropy-based 加权等最简 baseline 对比、其他多教师 KD 方法的实验对比，且部分提升在标准差范围内。
  • Full review: Claude Code 全文七公理审稿

On-device speech emotion recognition (SER) is critical for real-time applications, yet large self-supervised models that excel at SER are too costly for edge devices. Multi-teacher knowledge distillation can compress them into a lightweight student, but two challenges remain: teacher reliability varies across batches, and logit-level distillation ignores inter-sample relational structure. We propose Adaptive Multi-teacher Relational Distillation (AMRD) to address both. A one-class SVM on each teacher’s logit similarity matrix assigns per-batch weights favoring more coherent teachers. A relational distillation loss aligns teacher and student similarity matrices, capturing structure that logit matching misses. On IEMOCAP and CREMA-D datasets across four student architectures, AMRD outperforms single-teacher distillation baselines in most settings, and ablations confirm both components yield complementary gains.

12. Evaluation of forced alignment of code-mixed speech: the case of Hindi-English

Authors: Ayushi Pandey, Pamir Gogoi, Kevin Tang

Categories: cs.CL Score: 5.47/10 (Obj:6 Id:4 Ind:8 Comp:5 Eff:5 Nov:3 Rep:8)

  • Strength: 句级混合训练声学模型达 4.15ms 平均边界误差，比单语替代方案低 10 倍；代码开源，可复现性良好。
  • Weakness: 使用过时的 MFA v1.0 且未与 MFA v3 或神经对齐器比较；实验 2 训练数据类型与切分粒度混淆，因果归因不可靠；核心方法是 2020 年同作者论文的直接增量，无新机制。
  • Full review: Claude Code 全文七公理审稿

Code-mixed speech poses unique challenges to forced alignment: expanded inventories, orthographic errors, and speaker variation. We evaluate forced alignment of Hindi-English code-mixed speech using the Montreal Forced Aligner. We address 2 problems: (1) free variation involving native vs non-native pairs and (2) phonemic boundary detection for mid-utterance English words. Bootstrapping strategies substantially outperform unmodified lexicons. Acoustic models trained on sentence-level code-mixed data achieve a mean error of 4.15ms, ie. ten times lower than monolingual Hindi (38.18ms) or isolated English (37.58ms) alternatives. Principled lexicon design and code-mixed training data are both essential for reliable alignment of bilingual speech.

13. LLM4OSC: Profile-Bound Natural Language Control with Deterministic Validation for Open Sound Control

Authors: Yuan-Yi Fan

Categories: cs.HC, cs.SD, eess.AS Score: 4.95/10 (Obj:5 Id:5 Ind:5 Comp:8 Eff:3 Nov:5 Rep:8)

  • Strength: 提出 wrong-send rate 指标和 propose-validate-send 范式，在开源 MIT 包中实现完整的 NL→OSC 安全架构，B0 规则后端以 ~0.05ms 达到 100% semantic accuracy / 0% wrong-send。
  • Weakness: 仅 20 个测试用例（1 设备、12 patterns）验证，LLM 后端 latency 3-4s 不可用于 show-critical 场景，核心结论是符号 policy 完全主导 LLM 贡献——LLM 在当前系统中无增量价值。
  • Full review: Claude Code 全文七公理审稿

Open Sound Control (OSC) is the dominant wire protocol for real-time parametric control in professional audio, live performance, and virtual production. Large language models can emit plausible OSC, but they hallucinate addresses, mishandle type tags, and fail under paraphrase- unacceptable in show-critical contexts. We present LLM4OSC, a local-first architecture in which models propose structured intent JSON over a human-reviewed device profile, and deterministic code validates, clamps, and encodes before any UDP send. We introduce a frozen evaluation harness with CI gates on wrong-send rate: mismatches that would still pass validation and transmit. On a Max/MSP hero profile (12 patterns; 8 literal + 8 paraphrase + 4 refusal cases), after profile tag enrichment, symbolic slot fill, NL refine, and a retrieval confidence gate, backends B0–B3 all pass frozen gates (100% semantic accuracy, 0% wrong-send). B0 (rules) remains the production default at ~0.05ms; LLM backends remain ~3-4s. Historical few-shot B2 accuracy of 62.5% rises to 100% on this suite only after symbolic post-processing- not because the 0.5B model alone becomes show-safe. We argue for propose-validate-send and wrong-send rate as first-class metrics for language-to-control systems.

14. Device Invariance using Domain Adaptation on Acoustic Scene Classification

Authors: Abhishek dileep, Shubham Sharma, Padmanabhan Rajan

Categories: eess.AS, cs.AI, cs.SD | 6 pages , 5 figures Score: 4.84/10 (Obj:8 Id:5 Ind:8 Comp:5 Eff:5 Nov:3 Rep:6)

  • Strength: 系统比较了 DANN 和 CDAN 在 CNN 与 transformer 特征提取器上的表现，发现 DANN 在两种架构上均有效，而 CDAN 在 PaSST transformer 上发生模式崩溃——这是一个值得社区关注的实用警示。
  • Weakness: 因果归因缺乏控制实验（混淆了架构与预训练数据/模型规模），未与 DCASE 2020 Task 1A 排行榜或先前 DA 方法（Gharib 2018, Yang 2021, Olvera 2022, Tan 2024）进行比较，且 CDAN 失败未做诊断分析，仅有一个数据集和一个 transformer 实例，无法支持“方法需按特征表示定制”的普适结论。
  • Full review: Claude Code 全文七公理审稿

This paper explores the effectiveness of domain adaptation techniques when using convolutional neural network (CNN)-based and transformer-based feature representations for acoustic scene classification. Two well-known domain adaptation techniques, namely domain adversarial neural network (also called DANN) and conditional domain adversarial network (also called CDAN) are evaluated under various domain shifts. Our study indicates that DANN provides effective domain adaptation fairly consistently for both feature extractors. On the other hand, CDAN provides effective domain adaptation only for CNN-based feature extractors. The study gives insights into how domain adaptation methods may need to be tailored to the underlying feature representation. Experimental evaluation with multiple devices on the DCASE 2020 dataset supports the observations.

15. A Cross-lingual Comparison of Human and Classification Model Entrainment Behavior in Code-switched Speech Settings

Authors: Debasmita Bhattacharya, Siying Ding, Alayna Nguyen, Julia Hirschberg

Categories: cs.CL Score: 4.84/10 (Obj:6 Id:4 Ind:5 Comp:5 Eff:5 Nov:5 Rep:6)

  • Strength: 首次跨语言（普通话/印地语/西班牙语-英语）验证口语 CSW entrainment，发现 lexical entrainment 跨语言泛化、acoustic-prosodic/CSW style 因语言而异，并释放带 CSW style 标注的 MaSaC 语料。
  • Weakness: RQ2 存在标签-评测闭环循环论证，缺少 random/最简 baseline 和完整 accuracy 对照表，MaSaC 脚本化语料使印地语-英语结论可能反映语料而非语言学属性，且关键解释（M-index ceiling）未经独立验证。
  • Full review: Claude Code 全文七公理审稿

Conversational entrainment is well-studied in monolingual and written contexts, but remains underexplored in spoken code-switching (CSW). We present a novel cross-lingual analysis of entrainment in Mandarin-English, Hindi-English, and Spanish-English dialogue and show that, while lexical entrainment generalizes across language pairs, entrainment over acoustic-prosodic and CSW style aspects exhibits context-specific variation. We build on these findings by asking whether classification models capture these human behavioral patterns. Applying feature importance and ablation analyses, we find that classical and Transformer-based classifiers detect entrainment reasonably well but consistently prioritize features other than those most salient to human entraining behavior. Our approach introduces a human-grounded framework for evaluating model decision-making in multilingual stylistic contexts, and suggests future challenges for developing conversational agents capable of producing naturalistic code-switched speech.

16. SpeechLLM Meets Federated Learning for End-to-End ASR: English and Italian Case Studies

Authors: Mohamed Nabih Ali, Daniele Falavigna, Alessio Brutti

Categories: cs.CL | Accepted at the 2nd International Conference on Federated Learning and Intelligent Computing Systems (FLICS2026) Score: 4.80/10 (Obj:6 Id:5 Ind:8 Comp:3 Eff:5 Nov:3 Rep:8)

  • Strength: 在 PEFT-only 聚合设置下，联邦 SpeechLLM 在 LibriSpeech-100 英语单语场景达到 6.4% WER，接近 centralized 6.1%，证明 LoRA+projector 联邦训练可行且 gap 小（0.3%）。
  • Weakness: 核心方法”Adaptive FedAvg”仅为 FedAvg + 标准指数学习率衰减的命名包装，无新算法或机制；绝对 WER（6.1%）远弱于 SLAM-ASR SOTA（2.28%）；未比较 2024 年多篇直接相关的 FL+ASR PEFT 工作；通信成本”reduction”从未量化。
  • Full review: Claude Code 全文七公理审稿

Federated learning (FL) enables privacy-preserving training of automatic speech recognition (ASR) systems across distributed data sources, yet its application to large-scale speech language models (SpeechLLMs) remains unexplored. This paper presents the first systematic study of federated training for SpeechLLM-based end-to-end ASR systems. We design a communication-efficient federated optimization strategy tailored to the unique challenges of SpeechLLM architectures, addressing high-dimensional parameter spaces, gradient communication overhead, and computational constraints in distributed settings. Through extensive empirical evaluation on monolingual ASR tasks in English and Italian, we demonstrate the effectiveness and stability of our federated approach compared to centralized training baselines across diverse acoustic conditions and speaking styles. Additionally, we conduct a comprehensive ablation study analyzing the impact of different speech encoder architectures on monolingual English ASR performance within the federated framework, providing insights into optimal model configurations for decentralized training. Our results achieve competitive word error rates while reducing communication costs, establishing practical foundations for federated SpeechLLM deployment in real-world multilingual scenarios.