论文解读
口吃语音先转文字再修语义：STEAMROLLER 为何用多智能体绕开直接语音转换
针对口吃语音直接转流利语音难、转写引入语义失真和实时延迟三难，STEAMROLLER 用转写加多智能体文本修复加音色克隆合成的三段管线，在 FluencyBank 上把词错误率平均降约 10 个百分点并保持语义，代价是约 3.8 秒分段延迟和零样本克隆音色相似度偏低。
模型集成
论文解读
证据分散且多解：用分角色辩论与淘汰赛留下更站得住的关系判断
针对从双人对话推断熟人与陌生人及具体熟人关系的问题，论文提出免训练的多角色辩论与竞争裁决流程，在同一过滤测试集上比较零样本与辩论基线，结构化交互在文本与音文结合上取得宏 F1 领先，但纯音频提升不稳定且多次调用带来更高推理成本。
论文解读
猜对不算懂：用选项重排伪集成让音乐问答模型学会拒答
针对音乐音频语言模型在四选一问答中被迫作答而无法表达不知道的问题，论文用同一 TinyMU 检查点经答案不变扰动构造伪集成，主证据是四种选项排序平均将准确率从 55.7% 提升到 59.2% 并将误差保留曲线面积从 0.293 降到 0.261，代价是每次提问需要 4 次前向传播且无需重新训练。
论文解读
孤独听得出来吗：当说什么比怎么说更诚实
语音情感识别 | 4.9/10
论文解读
朗读课文不该读出抑郁：用证据边界把筛查关进可审计的笼子
语音情感识别 | 7.0/10
论文解读
Motion-Aware Reasoning from Speech to Mask Tracks: Runner-up Solution for the MeViS-Audio Track of the 8th LSVOS Challenge 2026
音视频理解 | 7.1/10
论文解读
Frame-Level Pansori Mode Classification with Complementary Audio Representations
音乐理解 | 7.6/10
论文解读
Masked diffusion enables coherent beat tracking
音乐理解 | 6.5/10
论文解读
Transfer Learning for Avian Bioacoustics under Sparse Positive Labels
音频分类 | 7.1/10
论文解读
Multi-Backbone Self-Supervised Ensembles for Audio Deepfake Detection and a Cross-Track Analysis of Generation-Detection Asymmetry
音频伪造检测 | 7.2/10
论文解读
Towards Operational Conversational Intelligence: A Speech Intelligence Framework
说话人日志 | 6.4/10
论文解读
Layer-Wise Decision Fusion for Fake Audio Detection Using XLS-R
音频理解 | 7.5/10
论文解读
AMECxSV: Adaptive Metadata-Driven Embedding-Fusion Calibration for X-Lingual Speaker Verification
说话人验证 | 7.8/10
论文解读
Component-Level Ensemble Fusion for Speech and Environmental Sound Deepfake Detection
语音伪造检测 | 6.4/10
论文解读
Team RAS in 11th ABAW Competition: Multimodal Ambivalence Recognition Approach
语音情感识别 | 5.3/10
论文解读
Can Tokens Compete? Token Representations against Supervised CNN Backbones for BirdCLEF+ 2026
音频事件检测 | 8.3/10
论文解读
HSEmotion Team at the 11th ABAW Challenge: Multi-Task Learning and Ambivalence/Hesitancy Video Recognition
音视频 | 7.9/10
论文解读
Simple Features and Honest Calibration for Ambivalence and Hesitancy Recognition in Video
模型集成 | 9.0/10
论文解读
Best-of-N TTS Evaluation is Confounded by ASR Family Alignment
语音质量评估 | 6.7/10
论文解读
SHAP-Weighted Cross-Modal Expert Fusion for Emotion and Sentiment Recognition: Evidence and Limits
语音情感识别 | 7.7/10