New forensic method uses AI sound decay to detect generated audio

Researchers have developed a new method to distinguish AI-generated impulsive sounds from real ones by analyzing group delay in the decay region. While onset-region group delays are similar, the decay region shows distinct patterns, with KL divergence reaching 0.322 for AI-generated sounds compared to 0.022 for real sounds. A Random Forest model utilizing nine decay-region features achieved an AUC of 0.884, and CNN classifiers using group delay maps reached 90-94% accuracy, indicating the potential of this technique as a forensic cue. AI

IMPACT This research introduces a new forensic technique for detecting AI-generated audio, potentially impacting content authenticity verification.

RANK_REASON Academic paper detailing a novel research method. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• AI-Generated Impulsive Sounds
• arXiv
• CNN
• Decay-Region Group Delay
• Kullback–Leibler divergence
• random forest
• short-time Fourier transform

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →