每日语音速递微信公众号：arXiv_Daily
cs.SD语音
标题：F3-Tokenizer：驯服音频自动编码器的理解和生成潜力
链接：https://arxiv.org/abs/2606.06357
备注：Technical report; early work; 9 pages, 2 figures, 5 tables
【2】FiLM-Based Speaker Conditioning of a SpeechLLM for Pathological Speech Recognition
标题：基于FiLM的语音说话人条件反射LLM用于病理性语音识别链接：https://arxiv.org/abs/2606.06211
备注：Accepted in Odyssey 2026: The Speaker and Language Recognition Workshop
【3】Learning Emotion-discriminative Representations for Zero-Shot Cross-lingual Speech Emotion Recognition
标题：Zero-Shot跨语言语音情感识别的描述区分表示链接：https://arxiv.org/abs/2606.06200
备注：Accepted to Interspeech 2026
【4】Multi-task Learning is Not Enough: Representational Entanglement in Dual-output Second Language Speech Recognition
标题：多任务学习还不够：双输出第二语言语音识别中的代表性纠缠链接：https://arxiv.org/abs/2606.06065
备注：5 pages, 2 figures, Accepted to the 43rd International Conference on Machine Learning Workshop on Machine Learning for Audio
【5】SpeechJBB: Probing Safety Alignment and Comprehension in Large Audio Language Models under Code-Switched Speech
标题：SpeechJBB：探索代码交换语音下大型音频语言模型的安全一致和理解链接：https://arxiv.org/abs/2606.06037
【6】DBHN-Net: Dual-Branch Hybrid Neural Network For Low-Complexity Monaural Speech Enhancement
标题：DBHN-Net：用于低复杂度单耳语音增强的双分支混合神经网络链接：https://arxiv.org/abs/2606.05911
备注：This article has been accepted for publication in IEEE Transactions on Pattern Analysis and Machine Intelligence(TPAMI)
【7】Beyond WER: A Paired Acoustic Stress Test for Ambient Clinical Scribes
标题：超越WER：环境临床抄写员的配对声学压力测试链接：https://arxiv.org/abs/2606.05909
备注：Accepted to INTERSPEECH 2026
【8】GLASS: GRPO-Trained LoRA for Acoustic Style Steering in Zero-Shot Text-to-Speech
标题：GLASS：接受GRPO训练的LoRA，用于Zero-Shot文本到语音中的声学风格引导链接：https://arxiv.org/abs/2606.05889
【9】UniVoice: A Unified Model for Speech and Singing Voice Generation
标题：UniVoice：语音和歌声生成的统一模型链接：https://arxiv.org/abs/2606.05852
备注：9 pages, 2 figures
【10】SagnacAssisted Enhanced OTDR for Distributed Acoustic Sensing: A Standardized Benchmark and Engineering Evaluation Framework
标题：用于分布式声学传感的Sagnac辅助增强型TLR：标准化基准和工程评估框架链接：https://arxiv.org/abs/2606.05754
【11】Do speech foundation models perceive speaker similarity as humans do?
标题：语音基础模型是否像人类一样感知说话者的相似性？链接：https://arxiv.org/abs/2606.05739
备注：Accepted by INTERSPEECH 2026
【12】Beyond Generative Decoding: Discriminative Hidden-State Readout from a Native Omni-Modal LLM for Multimodal Sentiment Analysis
标题：超越生成解码：用于多模式情绪分析的本地全模式LLM的区分性隐藏状态读出链接：https://arxiv.org/abs/2606.05713
备注：18 pages, 4 figures, 6 tables
【13】Beyond Waveform Robustness: Robust Feature-Vocoder Adversarial Attacks on Automatic Speech Recognition
标题：超越波形鲁棒性：对自动语音识别的鲁棒性干扰声码器对抗攻击链接：https://arxiv.org/abs/2606.05678
备注：11 pages
【14】SB-RF: Schrödinger Bridge Rectified Flow for One-Step Robust Speech Enhancement
标题：SB-RF：用于一步稳健语音增强的薛定汉桥纠正流程链接：https://arxiv.org/abs/2606.05575
【15】Sound Effects Dataset Unification With the Universal Category System
标题：音效数据集与通用类别系统的统一链接：https://arxiv.org/abs/2606.05571
备注：DAFx 2026 camera-ready version
【16】Domain-Aware Mispronunciation Detection and Diagnosis Using Language-Specific Statistical Graphs
标题：使用特定于语言的统计图的领域感知发音错误检测和诊断链接：https://arxiv.org/abs/2606.05569
备注：Accepted at Interspeech 2026
【17】Probing Spatial Structure in Pretrained Audio Representations
标题：探索预训练音频表示中的空间结构链接：https://arxiv.org/abs/2606.05544
备注：Accepted to Interspeech 2026
【18】Exploring LLMs for South Asian Music Understanding and Generation
标题：探索南亚音乐理解和生成的法学硕士链接：https://arxiv.org/abs/2606.05522
备注：19 pages, 7 figures
【19】nnAudio 2: Overcoming Dynamic Compilation Barriers and Transform Inconsistencies
标题：nnAudio 2：克服动态编译障碍并转变不确定性链接：https://arxiv.org/abs/2606.05394
【20】Task-Vector Arithmetic for Emotional Expressivity Control in Language-Model-Based Text-to-Speech
标题：基于语音模型的文本到语音中情绪表达力控制的任务-载体算法链接：https://arxiv.org/abs/2606.05367
备注：10 pages, 5 figures
【21】USAD 2.0: Scaling Representation Distillation for Universal Audio Understanding
标题：USAD 2.0：通用音频理解的缩放表示蒸馏链接：https://arxiv.org/abs/2606.06444
备注：Accepted to Interspeech 2026
【22】M2S-AVSR: Modality-aware Multi-view Self-supervised Representation for Robust Audio-Visual Speech Recognition
标题：M2 S-AVSR：用于鲁棒视听语音识别的模式感知多视图自我监督表示链接：https://arxiv.org/abs/2606.05763
备注：submitted to IEEE Transactions on Audio, Speech, and Language Processing
标题：USAD 2.0：通用音频理解的缩放表示蒸馏
链接：https://arxiv.org/abs/2606.06444
备注：Accepted to Interspeech 2026
【2】Revisiting Lexicon Evaluation in Unsupervised Word Discovery
标题：重新审视无监督单词发现中的词典评估链接：https://arxiv.org/abs/2606.06183
备注：6 figures
【3】CoSTA: Cognitive-State-Conditioned TTS Data Augmentation Using ASR Transcripts for Alzheimer's Disease Detection
标题：CoSTA：使用SVR转录物增强认知状态条件下的TTC数据以检测阿尔茨海默病链接：https://arxiv.org/abs/2606.06170
备注：Accepted by Interspeech 2026
【4】VoCodec: A Low-bitrate Streamable Neural Speech Codec with Voicing-driven Quantization
标题：VoCodec：具有语音驱动量化的低比特率可流传输神经语音编解码器链接：https://arxiv.org/abs/2606.05892
备注：Accepted to INTERSPEECH 2026
【5】An Ultra-Low-Bitrate Neural Speech Codec with Plain-to-Pseudo Synergistic Vector Quantization
标题：具有纯伪协同量化的超低比特率神经语音编解码器链接：https://arxiv.org/abs/2606.05876
备注：Accepted to INTERSPEECH 2026
【6】M2S-AVSR: Modality-aware Multi-view Self-supervised Representation for Robust Audio-Visual Speech Recognition
标题：M2 S-AVSR：用于鲁棒视听语音识别的模式感知多视图自我监督表示链接：https://arxiv.org/abs/2606.05763
备注：submitted to IEEE Transactions on Audio, Speech, and Language Processing
【7】Enhancing Audio Captioning with Auxiliary AudioSet Semantics
标题：使用辅助AudioSet语义增强音频字幕链接：https://arxiv.org/abs/2606.05717
【8】Age-Aware Adapter Tuning for Children's Speech Recognition
标题：儿童语音识别的感知适配器调整链接：https://arxiv.org/abs/2606.05440
备注：Our code is available at this https URL
【9】F3-Tokenizer: Taming Audio Autoencoder Latents for Understanding and Generation
标题：F3-Tokenizer：驯服音频自动编码器的理解和生成潜力链接：https://arxiv.org/abs/2606.06357
备注：Technical report; early work; 9 pages, 2 figures, 5 tables
【10】FiLM-Based Speaker Conditioning of a SpeechLLM for Pathological Speech Recognition
标题：基于FiLM的语音说话人条件反射LLM用于病理性语音识别链接：https://arxiv.org/abs/2606.06211
备注：Accepted in Odyssey 2026: The Speaker and Language Recognition Workshop
【11】Learning Emotion-discriminative Representations for Zero-Shot Cross-lingual Speech Emotion Recognition
标题：Zero-Shot跨语言语音情感识别的描述区分表示链接：https://arxiv.org/abs/2606.06200
备注：Accepted to Interspeech 2026
【12】Multi-task Learning is Not Enough: Representational Entanglement in Dual-output Second Language Speech Recognition
标题：多任务学习还不够：双输出第二语言语音识别中的代表性纠缠链接：https://arxiv.org/abs/2606.06065
备注：5 pages, 2 figures, Accepted to the 43rd International Conference on Machine Learning Workshop on Machine Learning for Audio
【13】SpeechJBB: Probing Safety Alignment and Comprehension in Large Audio Language Models under Code-Switched Speech
标题：SpeechJBB：探索代码交换语音下大型音频语言模型的安全一致和理解链接：https://arxiv.org/abs/2606.06037
【14】To Be Multimodal or Not to Be: Query-Adaptive Audio-Visual Person Retrieval via Active Modality Detection
标题：多模式还是非模式：通过主动模式检测的查询自适应视听人检索链接：https://arxiv.org/abs/2606.05931
备注：INTERSPEECH 2026
【15】DBHN-Net: Dual-Branch Hybrid Neural Network For Low-Complexity Monaural Speech Enhancement
标题：DBHN-Net：用于低复杂度单耳语音增强的双分支混合神经网络链接：https://arxiv.org/abs/2606.05911
备注：This article has been accepted for publication in IEEE Transactions on Pattern Analysis and Machine Intelligence(TPAMI)
【16】Beyond WER: A Paired Acoustic Stress Test for Ambient Clinical Scribes
标题：超越WER：环境临床抄写员的配对声学压力测试链接：https://arxiv.org/abs/2606.05909
备注：Accepted to INTERSPEECH 2026
【17】GLASS: GRPO-Trained LoRA for Acoustic Style Steering in Zero-Shot Text-to-Speech
标题：GLASS：接受GRPO训练的LoRA，用于Zero-Shot文本到语音中的声学风格引导链接：https://arxiv.org/abs/2606.05889
【18】UniVoice: A Unified Model for Speech and Singing Voice Generation
标题：UniVoice：语音和歌声生成的统一模型链接：https://arxiv.org/abs/2606.05852
备注：9 pages, 2 figures
【19】Towards Truly Multilingual ASR: Generalizing Code-Switching ASR to Unseen Language Pairs
标题：迈向真正的多语言ASB：将代码转换ASB推广到看不见的语言对链接：https://arxiv.org/abs/2606.05846
备注：ICML 2026 Workshop on Machine Learning for Audio
【20】FORTE: FOL-guided Optimal Refinement for Text-audio rEtrieval
标题：FOTE：文本音频rEtrieval的follow引导的最佳细化链接：https://arxiv.org/abs/2606.05812
备注：Under Review
【21】SagnacAssisted Enhanced OTDR for Distributed Acoustic Sensing: A Standardized Benchmark and Engineering Evaluation Framework
标题：用于分布式声学传感的Sagnac辅助增强型TLR：标准化基准和工程评估框架链接：https://arxiv.org/abs/2606.05754
【22】Do speech foundation models perceive speaker similarity as humans do?
标题：语音基础模型是否像人类一样感知说话者的相似性？链接：https://arxiv.org/abs/2606.05739
备注：Accepted by INTERSPEECH 2026
【23】Beyond Generative Decoding: Discriminative Hidden-State Readout from a Native Omni-Modal LLM for Multimodal Sentiment Analysis
标题：超越生成解码：用于多模式情绪分析的本地全模式LLM的区分性隐藏状态读出链接：https://arxiv.org/abs/2606.05713
备注：18 pages, 4 figures, 6 tables
【24】SB-RF: Schrödinger Bridge Rectified Flow for One-Step Robust Speech Enhancement
标题：SB-RF：用于一步稳健语音增强的薛定汉桥纠正流程链接：https://arxiv.org/abs/2606.05575
【25】Sound Effects Dataset Unification With the Universal Category System
标题：音效数据集与通用类别系统的统一链接：https://arxiv.org/abs/2606.05571
备注：DAFx 2026 camera-ready version
【26】Domain-Aware Mispronunciation Detection and Diagnosis Using Language-Specific Statistical Graphs
标题：使用特定于语言的统计图的领域感知发音错误检测和诊断链接：https://arxiv.org/abs/2606.05569
备注：Accepted at Interspeech 2026
【27】Probing Spatial Structure in Pretrained Audio Representations
标题：探索预训练音频表示中的空间结构链接：https://arxiv.org/abs/2606.05544
备注：Accepted to Interspeech 2026
【28】Exploring LLMs for South Asian Music Understanding and Generation
标题：探索南亚音乐理解和生成的法学硕士链接：https://arxiv.org/abs/2606.05522
备注：19 pages, 7 figures
【29】nnAudio 2: Overcoming Dynamic Compilation Barriers and Transform Inconsistencies
标题：nnAudio 2：克服动态编译障碍并转变不确定性链接：https://arxiv.org/abs/2606.05394
【30】Task-Vector Arithmetic for Emotional Expressivity Control in Language-Model-Based Text-to-Speech
标题：基于语音模型的文本到语音中情绪表达力控制的任务-载体算法链接：https://arxiv.org/abs/2606.05367
备注：10 pages, 5 figures
【31】MCBench: A Multicontext Safety Assessment Benchmark for Omni Large Language Models
标题：MCBench：Omni大型语言模型的多上下文安全评估基准链接：https://arxiv.org/abs/2606.05177
机器翻译由腾讯交互翻译提供，仅供参考
