For most of history, when we heard a voice, we could trust that a human was speaking. Generative AI has broken that trust. Synthetic voices now pass for real ones in recordings, phone calls, and voice commands. The standard response to this problem has been to train AI models to detect fakes, but generative models are evolving so quickly that the detectors can’t keep up. OriginStory takes a different approach: instead of asking whether content is fake, it proves that a real human created it, at the moment it was recorded.
In this session, OriginStory co-founder Bella Lenz walks through the edge architecture behind the platform. A microphone captures the speech, and a radar sensor acts as a witness to the human process that produces it. Human speech is generated by physiological motion: vocal fold vibration, breathing, heartbeat, and the movement of the jaw and articulators. The radar captures those motions as they happen. When no human is present, those signals are missing. OriginStory’s algorithm is trained on the correlation between these physiological signals and the audio they produce, so it can confirm that a live human generated the speech. The result is cryptographically bound to the audio and signed at capture. Nothing goes to the cloud. Only the proof travels.
Bella will cover the radar sensing and signal-processing pipeline and the benefits of running verification on the edge. Attendees will leave with an understanding of why verifying humans at the origin matters, and how sensors and edge compute make it possible.