音訊與語音模型
中級1 分鐘閱讀更新於 2026-03-15
語音辨識與音訊處理 AI 的安全——涵蓋對抗性音訊、語音複製風險、隱藏命令與音訊注入技術。
語音與音訊處理 AI 引入文字處理不存在的獨特攻擊面。對抗性音訊可嵌入人類聽不到但 AI 識別為命令的訊號。語音複製可冒充授權使用者。隱藏命令可觸發 AI 代理採取未授權動作。
學習路徑
0/5 已完成~44 分鐘總計5 課
- 17mSpeech Recognition Attacks進階
Attacking automatic speech recognition systems including adversarial audio that transcribes differently than heard, hidden voice commands, and background audio injection.
- 27mAdversarial Audio專家
Techniques for crafting adversarial audio perturbations including psychoacoustic hiding, frequency domain attacks, and over-the-air adversarial audio.
- 38mVoice Cloning進階
Voice cloning for social engineering against AI systems, voice authentication bypass, speaker verification attacks, and detection techniques.
- 47mLab: Audio Adversarial進階
Hands-on lab creating adversarial audio examples using Python audio processing, targeting Whisper transcription with injected text.
- 515mAudio Modality Attacks進階
Comprehensive attack taxonomy for audio-enabled LLMs: adversarial audio generation, voice-based prompt injection, cross-modal split attacks, and ultrasonic perturbations.