New framework uses spectro-temporal modulation for human-imitated speech detection

Researchers have developed a new Spectro-Temporal Modulation (STM) representation framework to better detect human-imitated speech. This approach uses cochlear filterbank models to capture both temporal and spectral fluctuations in speech, mimicking human auditory perception. Experiments show that STM representations, particularly Segmental-STM, are highly effective, even surpassing human performance in distinguishing imitated speech from genuine audio. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Introduces a novel method for detecting sophisticated voice imitation, potentially enhancing security in voice authentication systems.

RANK_REASON Academic paper proposing a novel framework for speech detection.

Read on arXiv cs.CL →

• Gammatone Filterbank
• Gammachirp Filterbank

• paper
• other