/ About this event
An evening of structured debate on the frontier of voice AI, hosted alongside SF Tech Week. Three sessions with the VP of Research from frontier Voice AI Labs, each built around a live debate on a hot, controversial, and a little provocative topic rather than a lecture.
Topic 1: End-to-end speech models vs. cascaded pipelines — which architecture wins? Voice AI systems are being built two fundamentally different ways. One approach processes speech end-to-end in a single model — faster, more natural, but harder to inspect and control. The other breaks the task into discrete stages — transcription, reasoning, synthesis — trading some fluency for traceability and control. Major labs and infrastructure companies are committing significant resources to opposing bets. This session examines which architecture holds up in production, and why the answer may differ by use case.
Topic 2: The data question — sourcing, scarcity, and the buy-vs-build decision Model quality in voice AI depends on real, naturally occurring conversation — not scripted or read speech. That data is scarce, expensive, and difficult to source at the quality frontier labs now require. This session addresses how leading teams are approaching data strategy: building proprietary collection pipelines versus licensing from external providers, and how long current sources of supply can be expected to meet demand.
Topic 3: The ethics and governance of synthetic voice As voice AI becomes indistinguishable from human speech, a set of governance questions has moved from theoretical to urgent: disclosure requirements when a user is speaking with AI, consent standards for the voices used in training, and the emerging legal and regulatory frameworks governing biometric voice data. This session examines where the industry's practices currently stand — and where they are likely headed.
This event is a part of #SFTechWeek—a week of events hosted by VCs and startups to bring together the tech ecosystem. Learn more at www.tech-week.com
