研究人员开发了一个名为 GRIDS 的新框架，用于分析扰动如何影响自监督语音模型的内部表示。通过使用局部内在维度 (LID)，该框架可以检测这些表示中的异常。研究发现，LID 升高与自动语音识别中的词错误率增加相关，从而能够进行无转录监控。 AI
影响 引入了一种检测语音模型异常的新颖方法，可能提高鲁棒性和安全性。
排序理由 学术论文，详细介绍了一种分析语音模型表示的新框架。
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
New GRIDS framework detects anomalies in self-supervised speech models
报道来源 [2]
-
Dimensionality-Aware Anomaly Detection in Learned Representations of Self-Supervised Speech Models
arXiv:2605.02715v1 Announce Type: cross Abstract: Self-supervised speech models (S3Ms) achieve strong downstream performance, yet their learned representations remain poorly understood under natural and adversarial perturbations. Prior studies rely on representation similarity or…
-
Dimensionality-Aware Anomaly Detection in Learned Representations of Self-Supervised Speech Models
Self-supervised speech models (S3Ms) achieve strong downstream performance, yet their learned representations remain poorly understood under natural and adversarial perturbations. Prior studies rely on representation similarity or global dimensionality, offering limited visibilit…