Florent Van Calster, PDG de Whispeak – 25/06

Florent Van Calster, CEO of Whispeak, opened the conversation with a striking claim: his company can detect a synthetic voice with 99% accuracy on benchmark datasets, outperforming major AI voice generators like ElevenLabs on their own detection tests. The episode quickly moved from this technical feat into the practical psychology of voice identity—how we trust what we hear, how that trust can be exploited, and what it takes to protect it.

Whispeak’s technology rests on two pillars. The first is voice biometrics: recognizing who is speaking by analyzing over 180 vocal factors—frequency, timbre, pauses, metallic undertones, and more. This is already in use in call centers, where a customer’s identity is verified passively within seconds of speaking, eliminating the friction of security questions, PINs, or two-factor authentication. The second pillar is detecting whether a voice is real or AI-generated. This is where the company claims world-leading performance, having won a defense AI competition with less than 1% error on a specific dataset. Van Calster noted that ElevenLabs’ own detector is currently at 0% accuracy against Whispeak’s system, though the goal is to approach 100% across all major voice synthesis engines.