Programme and participation
Voice Research Club #21 will take place on 3 September from 17:30 to 20:30 Pacific Daylight Time in San Francisco. The exact address is provided after an application is approved. Registration requires approval, capacity is limited and the event page does not list a public price. The organiser has confirmed the date and current registration status. [1 · Bay Area Frontier Research Club #21]
The meeting is intended for speech and audio researchers, voice-system developers and company founders. The programme includes NVIDIA's PersonaPlex for full-duplex speech with role and voice control, as well as the practical limitations of real-time voice models: latency, interruptions, memory, emotional nuance, reliability, cost and testing quality. [1 · Bay Area Frontier Research Club #21]
Why voice-system technology is not yet standardised
An impressive voice is becoming a reproducible feature in itself. More defensible value lies in an evaluation dataset for a specific use case: sales, support, education or healthcare. Recovery after interruptions, successful task completion, robustness to accents, contextual memory and the cost per conversation that actually resolves the task all matter. [1 · Bay Area Frontier Research Club #21]
A format centred on research papers allows discussion of systematic failure modes as well as the best demonstrations. For a commercial product, it is particularly important to test model behaviour in noise, overlapping speech, shifts in emotional tone and situations requiring correction of a misunderstood user intention. [1 · Bay Area Frontier Research Club #21]
Sources
- Bay Area Frontier Research Club #21 — 3 September 2026, 17:30–20:30 Pacific Daylight Time; approval required