论文解读
用可学习的分组有理激活替换固定激活：KanAdapter 在语音基座模型上的并行瓶颈适配
针对全量微调语音自监督基座模型代价高、MLP 适配器表达受限的问题，KanAdapter 以并行瓶颈中的 GR-KAN 可学习有理激活替换固定激活，在说话人验证、情感识别与伪造检测上以减少 94.7% 至 97.5% 可训练参数接近全量微调，并在两阶段持续学习上相对全量微调降低 83.6% 错误率。
说话人验证
论文解读
说话人一换就失准：用更强的判别器逼编码器减少说话人线索
针对未见说话人上情感识别易受说话人差异干扰的问题，论文用 wav2vec 2.0 做编码器、ECAPA-TDNN 做说话人判别器并以熵最大化做对抗，在 IEMOCAP 测试集报告 60.63% UA，代价是交替训练更复杂且存在原文内部数字不一致与未验证的跨库边界。
论文解读
复述一句就能听出你来过：微调语音如何记住嗓音与录音
语音合成 | 7.5/10
论文解读
当嵌入不再确定：把不确定性从打分一路带到校准的后端闭环
说话人验证 | 7.8/10
论文解读
Target Speaker Identification: A Low-Latency Streaming Pipeline
说话人验证 | 5.6/10
论文解读
In Defense of Using Worst-case Privacy Disclosure as Privacy Evaluation Metric of Voice Anonymization
说话人验证 | 7.1/10
论文解读
The Voiceprint Fallacy: Why Voices Are Not Unique Biometric Imprints
说话人验证 | 6.0/10
论文解读
Beyond Residual Connections: Manifold-Constrained Hyper-Connections for Robust Speaker Representation Learning
说话人验证 | 6.0/10
论文解读
Speaker Verification Under Real Classroom Conditions for English Speech
说话人验证 | 5.7/10
论文解读
Homebot: A Personal AI Agent for Conversational Home Assistance and Automation
语音交互 | 3.8/10
论文解读
QR-Erase: Efficient Subspace-Based Machine Unlearning with Layer Localization
说话人验证 | 6.0/10
论文解读
SCOPE: Entanglement Frontier Escape for Source-Free Class Unlearning
说话人验证 | 6.3/10
论文解读
Extracting Voice Styles from Frozen TTS Models via Gradient-Based Inverse Optimization
语音克隆 | 7.9/10
论文解读
Simple Language Normalization Wins: Cross-Lingual Speaker Verification for the TidyVoice 2026 Challenge
说话人验证 | 6.8/10
论文解读
Improving the performance of an ASV system using hybrid speech features
说话人验证 | 5.0/10
论文解读
Multimodal Speaker Verification as a Threat to Speaker Anonymization
说话人验证 | 9.2/10
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
Do Speech Tokens Leak Voiceprints? Speaker Inversion Attacks Against End-to-End Speech Language Models
说话人验证 | 7.7/10
论文解读
A Geometry-Limited Identification Floor and Its Consequences for Voice-Clone Attribution in Professional Voice Actors
说话人验证 | 8.8/10
论文解读
Rethinking Speech Foundation Model Fine-tuning: Better SFT or Better Match?
语音情感识别 | 6.7/10