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。

2026-07-09
日期2026-07-09
已评分—
均分—
最高—

Daily Papers — 2026-07-09

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

1. Best-of-$N$ TTS Evaluation is Confounded by ASR Family Alignment

Authors: Taehyung Yu, Seongjae Kang

Categories: cs.CL, cs.AI, cs.LG, cs.SD | Accepted at ICML 2026 Workshop on Machine Learning for Audio Score: 8.10/10 (Obj:9 Id:8 Ind:8 Comp:9 Eff:7 Nov:8 Rep:9)

  • Strength: 中 — 提出的跨家族 rank 集成方法简单有效，但方法本身是标准 rank aggregation 的新应用，技术复杂度有限。
  • Weakness: 绝对指标处于可用范围：F5-TTS 基线 WER 2.06%，最佳单验证器降至 1.72%（−16.5%），跨家族集成在 N=10 达到 1.61% mean WER（−12%）。改进在不同评估器下广泛存在：rank-avg 在 N=5 是唯一在三个评估器下同时达到 p<0.05 的配置（.001/.022/.0002）。音质指标（SIM-o, UTMOS）无退化。
  • Full review: Claude Code 全文七公理审稿

Best-of-$N$ (BoN) inference improves content consistency in zero-shot text-to-speech by selecting from $N$ candidates with an automatic speech recognition (ASR) verifier. We identify an underexplored evaluation confound: a verifier’s apparent quality depends strongly on which ASR family judges it. On LibriSpeech-PC test-clean~\citep{librispeechpc} with F5-TTS~\citep{f5tts}, verifier rankings reverse across Whisper, wav2vec~2.0, and HuBERT evaluators, and same-family verifier-evaluator pairs recover 2-3$\times$ more oracle headroom than cross-family pairs despite near-identical representations (linear CKA $0.978$) – a pattern consistent with identity- or lineage-level coupling rather than representational overlap. We propose two \textbf{cross-family rank ensembles} (rank-averaging and conjunctive max-rank) that attain the lowest mean WER across three independent evaluators – $1.61\%$ at $N{=}10$ ($-12\%$ relative to F5-TTS) – with no measurable degradation under automatic SIM-o/UTMOS metrics; the best single verifier drives WER from $2.06\%$ to $1.72\%$ ($-16.5\%$) under the official F5-TTS evaluator. We recommend cross-evaluator triangulation as default reporting practice.

2. Structural Bottlenecks on Frequency Representation in End-to-End Audio Models

Authors: Nicole Cosme-Clifford

Categories: cs.SD, cs.LG Score: 7.40/10 (Obj:9 Id:8 Ind:8 Comp:8 Eff:6 Nov:8 Rep:5)

  • Strength: 中 — GLRF 作为 post-hoc retraining-free 干预方法概念优雅且实用，但技术本质是 Gabor 基投影，创新度不算高。效用验证偏窄，未在主流 benchmark 上展示实际任务性能提升。
  • Weakness: 可复现性存在明显不足
  • Full review: Claude Code 全文七公理审稿

End-to-end neural audio models achieve high-fidelity compression and generation. We might read that performance as evidence they directly represent interpretable features such as pitch and timbre, but a model can produce plausible outputs without doing so. A model may encode these features in any reachable basis, but regardless of which, the features are well described as compositions of time-frequency-localized primitives. Whether state-of-the-art encoders preserve access to these primitives, and thus to compositions of them, remains unclear. Through theoretical analysis and controlled experiments, we show that several state-of-the-art strided convolutional encoders impose two structural bottlenecks, both predictable from architecture and signal structure, on access to these primitives: (1) they collapse primitives into alias equivalence classes, establishing a bound on representational capacity, and (2) they limit the frequency resolution available to learned filters, restricting separability. For well structured data, we find collapse rates of 31-35% and filter bandwidths 10-35x above the theoretical resolution bound, confirming that both bottlenecks arise under realistic signal conditions. We then introduce Gabor Latent Refactorization (GLRF), a lightweight post-hoc intervention that re-expresses encoder latents in a frequency-localized basis, reducing filter bandwidths from 10-35x to 1.5-3x of the theoretical resolution bound while preserving reconstruction fidelity and improving control over attributes like pitch. These results show that the encoders in question predictably degrade access to frequency-localized primitives, entangling the features that depend on them, and that a lightweight, retraining-free intervention can recover much of that access, improving steerability and interpretability.

3. MulTTiPop: A Multitrack Transcription Dataset for Pop Music

Authors: Nathan Pruyne, Benjamin Stoler, William Chen, Chien-yu Huang, Shinji Watanabe et al.

Categories: cs.SD, cs.LG | 8 pages, 4 figures. Associated web preview available at https://gclef-cmu.org/multtipop Score: 7.26/10 (Obj:9 Id:6 Ind:8 Comp:6 Eff:8 Nov:7 Rep:8)

  • Strength: Pipeline 是已有方法的合理组装，anchor beat 选择的消融分析有参考价值，但 49.1% 成功率表明方法效率有显著改进空间。
  • Weakness: 作为数据集论文，识别公理的核心问题是：数据构建方法中，哪些环节真正导致了高质量对齐，哪些环节是瓶颈？ 论文在 Section 3.1 提供了 anchor beat selection 的消融分析（base 2.3% → melody 31.7% → ytmelody 40.0%），这部分是有价值的因果识别。但存在以下缺口
  • Full review: Claude Code 全文七公理审稿

We present MulTTiPop, a dataset of pop music segments and their associated multitrack MIDI recordings for the evaluation of automatic music transcription models. MulTTiPop contains 572 segments of popular music totaling 3.5 hours of audio, and contains songs from diverse genres and decades from the 1930s to 2000s. To collect this dataset, we perform metadata-based matching on song segments from the Lakh MIDI and TheoryTab datasets, manually identify an anchor beat between the audio and MIDI, then use beat tracking on the audio and warp the MIDI to match its tempo and timing. We evaluate state-of-the-art automatic music transcription models on MulTTiPop and find substantial room for improvement, with the best model achieving 38% Onset F1. More details and sound examples of MulTTiPop are available at https://gclef-cmu.org/multtipop.

4. A Quantized Native Runtime for On-Device Semantic Audio Generation

Authors: Matteo Spanio, Antonio Rodà

Categories: cs.SD, cs.PF | Under review at International Symposium on the Internet of Sounds (IS2) Score: 7.05/10 (Obj:8 Id:7 Ind:9 Comp:7 Eff:8 Nov:5 Rep:9)

  • Strength: aria runtime (~7.7k LOC C/CUDA) achieves 7× faster cold start, warm parity, and runs the 1.2B-parameter SA3 model on an 8GB Raspberry Pi 5 at 4-bit, with 8-bit showing no measurable quality loss across three independent。
  • Weakness: Core novelty is a port of established techniques (ggml-style native runtime, post-training quantization, activation steering) to audio diffusion — no new quantization algorithm, steering mechanism, or runtime architectur。
  • Full review: Claude Code 全文七公理审稿

Semantic audio applications increasingly require controllable generation on commodity and embedded hardware rather than through framework-heavy datacenter stacks. We present \textit{aria}, a dependency-free native runtime that runs the complete text-to-music pipeline of Stable Audio~3 (SA3) on ordinary GPUs, CPU-only machines, and a Raspberry~Pi~5, with no Python or deep-learning framework underneath. Our main contribution is a study of quantization: running the model at lower numerical precision to fit tight memory budgets, saving memory in place rather than adding to it. Because the runtime owns every internal tensor, it also exposes activation steering, a low-cost way to steer what the model generates. We judge the quality cost with three independent measures of the output (prompt adherence, overall audio quality, taste preservation), each compared against the ordinary variation between random seeds. Eight-bit precision shows no measurable quality loss on any measure while sharply cutting memory, and it is the fastest mode on the GPU; four-bit adds a small, bounded cost but shrinks the footprint enough to run the $1.2$-billion-parameter model on an $8$\,GB Pi. Against the official implementation, aria matches or exceeds generation speed and starts about seven times faster. A case study of the steering interface generates music carrying taste associations (\emph{sonic seasoning}), with genuine but bounded control for a subset of attributes. These results make a compact, quantized runtime with built-in control a practical basis for on-device semantic audio in Internet-of-Sounds settings. The \textit{aria} runtime is released at https://github.com/matteospanio/aria.

5. When Synthetic Speech Is All You Have: Better Call GRPO

Authors: Shashi Kumar, Yanis Labrak, Hasindri Watawana, Sergio Burdisso, Esaú Villatoro-Tello et al.

Categories: cs.CL, cs.AI | Submitted to SLT 2026 Score: 6.50/10 (Obj:8 Id:8 Ind:6 Comp:8 Eff:6 Nov:6 Rep:7)

  • Strength: 中高 — 系统的六维实验设计（SFT vs GRPO, real data scaling, synthetic data scaling, reward ablation, selection strategy, mechanism analysis）和方法论发现（WER reward 足够, random sampling 足够, 5-10h real 是甜点）有实践价值。
  • Weakness: reward 与 evaluation 重叠: GRPO 的默认 reward 是 RWER = 1 - WER，而最终评估指标也是 WER。这是 reward 直接等于 evaluation metric 的情况。虽然论文在 Table IV 中消融了不同 reward（CER, LEN 组合），但主结论（GRPO 优于 SFT）基于的实验使用了与评估指标完全相同的 reward。
  • Full review: Claude Code 全文七公理审稿

LLM-based ASR adapted to regulated domains such as banking is bottlenecked by privacy: real speech is costly and legally constrained to collect, making synthetic text-to-speech (TTS) an attractive substitute. Yet synthetic speech stays acoustically mismatched with real recordings, and work on this gap has stayed within supervised fine-tuning (SFT). We instead turn to reinforcement learning, and show that Group Relative Policy Optimization (GRPO) extracts far more from the same synthetic speech than SFT. Synthetic-only adaptation of the model with GRPO, a critic-free method rewarding low-WER hypotheses, reduces WER by 40\% relative to SFT (36.71\%$\to$22.09\%), and an SFT-then-GRPO combination pushes this further to 45\%. We trace the gain to behavior rather than representation: GRPO reduces insertion errors by improving stopping calibration and speech-to-text alignment by better anchoring attention to audio, leaving early-layer representations intact. When synthetic speech is the main resource, reinforcement learning should be preferred over supervised fine-tuning.

6. PS4: Proxy-Supervised Joint Training for Real Target Speaker Extraction

Authors: Wanyi Ning, Wei Zhou, Yingpeng Li, Yinshang Guo, Haitao Qian et al.

Categories: cs.SD, cs.AI Score: 6.40/10 (Obj:9 Id:4 Ind:5 Comp:5 Eff:8 Nov:5 Rep:8)

  • Strength: 低-中 — 四个已有损失的标准组合，无架构创新、无新机制、无理论分析。方法可复现但不可推广到其他问题设置（缺乏对损失权重、组件必要性的系统研究）。
  • Weakness: 论文没有提供消融实验来隔离四个代理损失（LCE, LSIM, LVAD, LDNSMOS）各自的作用。读者无法判断：哪个损失贡献最大？哪个损失是必要的？是否存在冗余？四个损失是否有协同效应？论文声称”四项互补代理监督目标提供有效的训练信号”，但无任何实验支持”互补”这一 claim。
  • Full review: Claude Code 全文七公理审稿

Training target speaker extraction (TSE) models for real conversational mixtures remains challenging because large-scale training corpora and clean target speech for supervision are unavailable. We present PS4, a proxy-supervised training framework for TSE in real conversational mixtures, with two main contributions. First, we construct a large-scale corpus of 71,771 training samples derived from four public datasets, covering both Chinese and English scenarios. Each sample contains an overlapping speech mixture, per-speaker enrollment audio, a ground-truth transcript, and frame-level voice activity labels. Second, we propose a proxy-supervised joint training strategy that fine-tunes a BSRNN-based TSE model using four complementary differentiable objectives: ASR cross-entropy, speaker similarity, frame-level voice activity detection, and perceptual audio quality. Starting from a publicly available pre-trained checkpoint, only the BSRNN separator is updated during fine-tuning. On the REAL-T challenge leaderboard, PS4 ranks 2nd overall, achieving the best speaker similarity and timing F1 among all submitted systems.

7. VSRo-200: A Romanian Visual Speech Recognition Dataset for Studying Supervision and Multimodal Robustness

Authors: Iulia-Maria Udrea, Alexandra Diaconu, Bogdan Alexe

Categories: cs.CV Score: 5.95/10 (Obj:8 Id:5 Ind:6 Comp:4 Eff:5 Nov:5 Rep:5)

  • Strength: 首个罗马尼亚语句子级 VSR 数据集（200h），人工 vs 伪标签受控对比显示 200h 伪标签 WER 48.8% 逼近 100h 人工标注 WER 53.3%，LRRo 迁移 Acc@1 从 71.0%→95.0%。
  • Weakness: 方法层面完全复用 MultiVSR 架构和标准 entropy 融合，无新方法贡献；AVSR 噪声评测仅 16 分钟、OOD 每类 15 分钟样本量过小；代码和模型 checkpoint 未公开，长期可复现性存疑。
  • Full review: Claude Code 全文七公理审稿

We introduce VSRo-200, the first large-scale dataset for visual speech recognition (lip reading) in Romanian, comprising 200 hours of real-world podcast videos. All samples are annotated with pseudo-labels generated by a fine-tuned Romanian ASR model, while a subset of 100 hours is additionally transcribed by humans, enabling controlled analysis of supervision quality under a unified framework. Building on this dataset, we establish a benchmark for visual speech recognition in low-resource settings. We systematically study the impact of supervision quality, showing that while human annotations provide better performance at fixed data scales, pseudo-labels enable continued improvements through scalability. We further evaluate robustness under domain shift using curated out-of-distribution (OOD) test sets, and analyze audio-visual speech recognition (AVSR) under noisy conditions, where multimodal fusion significantly improves robustness compared to audio-only models. Finally, we demonstrate that representations learned on VSRo-200 transfer effectively to the LRRo benchmark for isolated word recognition, substantially outperforming previously reported results. Overall, VSRo-200 provides a new testbed for studying supervision, domain generalization, and multimodal fusion in low-resource visual speech recognition.

8. COALA: Robust Contextualized Speech-augmented Language Modeling for ASR via Contrastive Regularizer and Biasing Score Estimation

Authors: Jhih-Rong Guo, Bi-Cheng Yan, Tien-Hong Lo, Berlin Chen

Categories: cs.CL | Accepted at INTERSPEECH 2026 Score: 5.95/10 (Obj:8 Id:6 Ind:8 Comp:5 Eff:6 Nov:5 Rep:8)

  • Strength: DPD-Loss 将 biasing scoring 重构为 point-wise binary classification，在 LibriSpeech N=5000 时 Recall#20 达 99.09%/96.59%（clean/other），显著超越 CTC-Filter 和 K-Prompt 基线。
  • Weakness: 未与最直接的同类工作 BR-ASR (Interspeech 2025) 比较，N=5000 时 ASR 推理 OOM 导致最大 biasing list 规模下结果缺失，且 unbiased B-WER 落后 baselines 9-11%。
  • Full review: Claude Code 全文七公理审稿

Contextual biasing seeks to integrate external knowledge into automatic speech recognition (ASR) systems to accurately recognize domain-specific entities. In this paper, we propose COALA (Contextualized ASR Leveraging Biasing Scoring), a robust framework designed to enhance speech-augmented language models (SLMs) in complex multi-entity scenarios. Considering the inherent context-window limitations of SLMs, identifying relevant target entities from a large-scale biasing list is crucial for effective recognition. To this end, COALA maps SLM latent representations into a specialized discriminative space to quantify the matching intensity between audio segments and candidate entities. Furthermore, we address the training collapse in prior study when handling multi-target utterances-where multiple rare words co-occur. Experimental results on the LibriSpeech benchmark demonstrate that COALA consistently achieves superior contextual biasing performance across various biasing list scales.

9. It Takes Few to TANGO: A Quantized Distributed Model for Binaural Speech Enhancement

Authors: Zahra Benslimane, Pierre Chouteau, Martyna Poreba, Fabrice Auzanneau, Michal Szczepanski et al.

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

  • Strength: 系统证明混合神经-空间架构中空间滤波阶段有效补偿 INT8 量化误差，据此将 MN-TANGO 压缩至 4.65 MMAC/s / 0.177 MB 且最终 GEVD 输出 STOI/PESQ 与 FP32 几乎持平（差距 <0.02）。
  • Weakness: 方法三组件（QAT、ERB 压缩、grouped LSTM）均直接复用先前工作（FQSS、RT-TANGO、Gao 2018），缺少与纯神经量化 SE baseline（如 TinyLSTM）在同一评测协议下的直接对比，且评测仅覆盖 BinauRec 子集（1200 mixtures, 3 SNR, 2 噪声方位），泛化性验证不足。
  • Full review: Claude Code 全文七公理审稿

Neural network-based multichannel speech enhancement systems achieve strong enhancement performance, but their computational and memory requirements limit deployment on resource-constrained devices. This paper investigates low-precision inference for TANGO, a hybrid distributed binaural speech enhancement system combining neural mask estimation with spatial filtering. We evaluate post-training quantization and quantization-aware training for the neural components, and analyze how quantization errors in the mask estimators propagate through the downstream spatial filtering stage. Our analysis shows that, although quantization degrades intermediate mask estimates, the spatial filtering stage compensates for most quantization-induced errors. Leveraging this robustness, we simplify TANGO into MN-TANGO, reducing both model size and computational complexity while maintaining comparable final performance. By combining INT8 weight-and-activation quantization with ERB compression and grouped recurrent layers, the most compact MN-TANGO reaches 4.65 MMAC/s and 0.177 MB.

10. MuScriptor: An Open Model for Multi-Instrument Music Transcription

Authors: Simon Rouard, Michael Krause, Axel Roebel, Carl-Johann Simon-Gabriel, Alexandre Défossez

Categories: cs.SD, cs.LG Score: 5.79/10 (Obj:8 Id:5 Ind:5 Comp:8 Eff:6 Nov:5 Rep:8)

  • Strength: 开放权重多乐器 AMT 模型，1.3B 参数，三阶段训练（合成 pre-training → 真实 fine-tuning → GRPO RL），在自建测试集上 Multi F1 达 47.8，较 YourMT3+ 的 21.9 提升 +25.9，是首个公开可用且跨 genre 有效的多乐器转录模型。
  • Weakness: 未与 2025 AMT Challenge 冠军 MIROS 比较；测试集与训练集同源于 DReal 且 RL reward 与评测指标重叠构成中等循环论证；Multi F1 < 50 时声称模型”works on real-world music”有过承诺风险；RL 阶段缺少 supervised fine-tuning 控制组以隔离算法贡献与数据贡献。
  • Full review: Claude Code 全文七公理审稿

Existing methods for automatic music transcription are often limited to single-instrument recordings or fail on complex, real music mixes. Although previous work utilizes synthetic training data, the resulting models generalize poorly, leading to largely unusable transcription output in realistic, multi-instrument settings. In this work, we analyze the effectiveness of synthetic data for pre-training while combining it with fine-tuning on real music audio and post-training using reinforcement learning. We further introduce conditioning on instrument presence to customize transcriptions. Finally, we release MuScriptor, an open-weight multi-instrument music transcription model that works on real-world music recordings from across a diverse range of musical genres.

11. Why Do You Say It Like That? A Phoneme-Level Framework for Explainable Speech Deepfake Detection

Authors: Anna Taylor, Michele Panariello, Massimiliano Todisco, Chiara Galdi, Nicholas Evans et al.

Categories: eess.AS, cs.SD Score: 5.53/10 (Obj:8 Id:5 Ind:6 Comp:5 Eff:5 Nov:4 Rep:6)

  • Strength: 低-中 — 方法核心（Grad-CAM + phoneme alignment）与 PDSM (Gupta et al. 2024) 高度重叠，统计分析方法论本身是标准统计工具的应用，没有新算法贡献。
  • Weakness: 论文面临严重的新颖性挑战：(1) PDSM (Gupta et al., Interspeech 2024) 已经提出了将 saliency map 按音素边界离散化用于深度伪造检测解释的核心想法。本文的 Grad-CAM + phoneme alignment + aggregation 的 pipeline 与 PDSM 在概念上高度相似。
  • Full review: Claude Code 全文七公理审稿

As the accuracy of speech deepfake detection improves with the use of self-supervised representations such as wav2vec 2.0 and HuBERT, understanding why the speech is classified as bona fide or deepfake remains an open challenge. In pursuit of more trustworthy and interpretable artificial intelligence, we introduce a phoneme-level analysis framework that connects model predictions to measurable phonetic units. Our post-hoc explainability method is generally applicable to a variety of speech deepfake detection systems based on convolutional neural networks since it leverages Gradient-weighted Class Activation Mapping in conjunction with speech recognition to generate saliency maps aligned with phonemes and pauses. This pipeline reveals statistically significant attack- and speaker-dependent phonetic cues associated with spoofed speech in terms that humans can understand. Experiments using ASVspoof 5 show comparable detection performance to similar architectures while providing linguistic interpretations across speakers and spoofing conditions.

12. On the Role of Conversational Timing in Synthetic Training Data for ASR

Authors: Máté Gedeon, Péter Mihajlik

Categories: eess.AS, cs.AI, cs.SD Score: 3.95/10 (Obj:6 Id:5 Ind:9 Comp:5 Eff:3 Nov:4 Rep:5)

  • Strength: 通过控制变量分析揭示 overlap-gap trade-off：intrinsic timing statistics (overlap rate ρ=-0.645, gap tail mass ρ=0.656, BH-significant) 比 raw θ coordinates 预测 cpWER 更强 (R²=0.529 vs 0.280)，且在 phase-controlled partial correlation 中保持稳。
  • Weakness: 效用极弱：25 个配置无 seed variance，best eval cpWER 仅比 corpus reference 好 0.3% absolute，dev set 无改善 (p=0.846)，cpCER 不达 BH 显著性，且仅在一个匈牙利语 corpus、单一 ASR 架构上验证。
  • Full review: Claude Code 全文七公理审稿

Synthetic multi-speaker conversations are widely used to train conversational automatic speech recognition (ASR) systems, but it remains unclear which timing properties make simulated data most useful. This paper studies conversational timing as a controllable training variable rather than merely as a corpus statistic to be reproduced. We parameterize pause and overlap timing distributions with an exponential-tilting family estimated from multiple conversational corpora, and then explore the resulting four-dimensional parameter space with Latin hypercube sampling and multi-objective Bayesian optimization. Each sampled timing configuration is used to generate simulated training conversations, train an ASR system, and evaluate concatenated-permutation word and character error rates (cpWER and cpCER) on a Hungarian dialogue corpus. The results show that downstream ASR behavior is explained more directly by induced timing statistics than by raw simulator coordinates or corpus proximity. In particular, higher overlap exposure is associated with lower cpWER, whereas longer and more variable gaps are associated with higher cpWER; cpCER follows the same trend, but with weaker statistical support. Bayesian optimization yields modest aggregate improvements, but its main value is analytical: it produces controlled timing interventions that reveal an overlap–gap trade-off in simulated conversational training data. These findings suggest that realistic simulation should be complemented by task-relevant diagnostics of overlap, gap, and timing-variability profiles.

13. Diarization-Guided Qwen-ASR Adaptation for Multilingual Two-Speaker Conversational Speech

Authors: Hao Wu, RongQi Han, Zhen Wang, Wei Liang, Wei Xu

Categories: cs.CL | 4 main pages plus 1 page of reference Score: 3.68/10 (Obj:6 Id:5 Ind:3 Comp:2 Eff:5 Nov:2 Rep:4)

  • Strength: 在 MLC-SLM 2026 Challenge 上，通过 SFT + LoRA + GRPO 将 Qwen3-ASR-1.7B 的 tcpMER 从 30.53 改善至 23.70（开发集）/ 17.97（评估集），在 21 种多语言环境下优于 Whisper-large-v3 和官方基线。
  • Weakness: 系统是对未修改的现成组件（3D-Speaker, LoRA, GRPO）的直接工程组合，且其 GRPO-ASR 奖励设计重复了已有工作 (arXiv:2509.01939, 2025年9月)；RL 贡献微不足道 (0.08 tcpMER)，且训练奖励直接重叠于评估指标。
  • Full review: Claude Code 全文七公理审稿

This paper describes our self-designed system for Task 1 of the MLC-SLM 2026 Challenge for multilingual two-speaker conversational speech. The system combines a modular speaker diarization front end with a challenge-adapted Qwen3-ASR-1.7B recognizer. The diarization front end performs voice activity detection, subsegment generation, CAMPPlus speaker embedding extraction, two-speaker spectral clustering, and RTTM-based audio segmentation. The resulting speaker-attributed segments are grouped by language or region and decoded by the adapted ASR model. For ASR adaptation, we first perform supervised full fine-tuning on the official training data, then apply LoRA fine-tuning with synthetic speech generated by a three-pipeline TTS-based synthetic speech augmentation framework, and finally refine the model using GRPO reinforcement learning with rewards based on WER/CER and penalties for hallucination, repetition, and length deviation. On the official development set, the full system achieves an average tcpMER of 23.70, reducing the error rate by 6.83 absolute points relative to the released Qwen-ASR-1.7B performance. On the final evaluation set, the system achieves an average tcpMER of 17.97. Ablation results show that supervised fine-tuning provides the largest gain, while synthetic-speech LoRA adaptation and reinforcement learning further improve robustness.