论文解读
整句重复又帧内粘连：MelTrim 先按声音粗筛再按梯度细剪做语音分类剪枝
针对语音分类中整句之间与句子内部同时冗余的问题，MelTrim 先用声学特征聚类做整句粗筛，再用冻结判别模型梯度范数做帧级细剪，在极小子集上取得明显精度优势，但选择本身带来一次性开销且当前只验证单一声学模态。
说话人识别
会议总览
eacl-2026 论文深度解读
共收录 36 篇 eacl-2026 会议论文的 Reader 深度解读
论文解读
多人对话先找对说话人，再把文本的情绪理解教给声音和画面
该研究用音画同步选出说话人、用文本教师蒸馏音频与视觉图网络、再用分层注意力与组合损失做融合，在 MELD 达到 67.75% 加权 F1、在 IEMOCAP 达到 72.44% 加权 F1，代价是引入更多超参数与调参负担。
论文解读
不加噪声也能留后门：用整体起伏的频率与响度曲线触发说话人识别
针对闭集说话人识别的投毒后门问题，论文用时域上的频率调制与幅度调制构造全局平滑触发，在中毒率 0.05 等条件下报告数字与物理场景的高攻击成功率，同时保持良性准确率下降有限，但干净标签等更难设置的性能明显走弱。
论文解读
把“谁何时说了什么”一次生成：SpeakerLM 如何统一日志与识别并兼容注册条件
针对级联式说话人日志与识别中误差传递与重叠语音难处理的问题，SpeakerLM 用音频编码器加投影器接入大语言模型的端到端多模态方案，在约 7638.95 小时数据下主指标超越最强级联基线，但小数据与仿真失配时仍明显落后。
论文解读
AMR: Adaptive Modality Routing for Multimodal Polyglot Speaker Identification
说话人识别 | 7.8/10
论文解读
DG^VoiC: Speaker Clustering for Fraud Investigation under Real Call-Centre Conditions
说话人识别 | 5.7/10
论文解读
Dialogue to Detection: A Multimodal Hybrid NLP Pipeline for Insurance Fraud Detection
说话人识别 | 6/10
论文解读
VieSpeaker: A Large-Scale Vietnamese Speaker Recognition Dataset Beyond Visual Dependency
说话人识别 | 7.5/10
论文解读
ESPnet3: Infrastructure for Scalable Speech and Audio Research in the Foundation Model Era
语音识别 | 7.5/10
论文解读
LISE : Listenable Interpretable Speaker Embeddings
说话人验证 | 6.8/10
论文解读
Stabilizing Short Duration Speaker Verification through Neural Re-scoring with Hybrid Enrollment
说话人验证 | 7.9/10
论文解读
MaskedFOP: Polyglot Speaker Identification under Missing Visual Modality via Cascaded Graph Label Propagation
说话人识别 | 9.2/10
论文解读
Multimodal Speaker Identification in Classroom Environments
说话人识别 | 6/10
论文解读
Who Spoke When in Multi-Conversation: Target Speaker Tagging Task and Benchmark
说话人识别 | 8.6/10
论文解读
Missing-Token Prompted Reliability-Aware Fusion for Robust Polyglot Speaker Identification
说话人识别 | 8.6/10
论文解读
Speaker Group Encoding in Self-supervised Speech Recognition Models
语音识别 | 6.5/10
论文解读
G-MaP-SE: Guided Speech Enhancement via GMM-Based Prior Matching
语音增强 | 9.3/10
论文解读
A Large-Scale Per-Speaker Analysis of Re-identification Risk in Speech Anonymization
语音匿名化 | 7.1/10
论文解读
Do speech foundation models perceive speaker similarity as humans do?
说话人识别 | 6.3/10