Preprint measures automated voices in unwanted calls

A new preprint reports that machine-voiced openings accounted for at least 26.9% of 7,233 analyzed unwanted calls captured by an interactive voice honeypot over 66 days. The researchers used repeated-recording fingerprints, a commercial synthetic-speech detector and blinded listener checks. They found detector-labeled synthetic openings concentrated in lead-generation spam and said only 0.44% disclosed automation. The findings are based on the study's honeypot sample and the synthetic-speech detector showed inconsistent classifications for identical waveforms.