Voice AI executives say the category has not had its ChatGPT moment

PolyAI chief technology officer Shawn Wen and Otter chief marketing officer Alex Gay, speaking on a HumanX conference panel reported by TechCrunch, said voice AI has not yet had its "ChatGPT moment." Wen said full-duplex models that can speak while listening have arrived, and that the next challenge is making reasoning fast enough that replies arrive quickly and conversations feel natural.

Both identified speech recognition as a weak link: Wen said ASR models often miss keywords and break context, and Gay said transcription errors flow into downstream actions and erode user trust. On embodiment, Gay said meeting digital twins must convey human-like emotion, and that without it "then it's just a Q and A chatbot." Wen also argued enterprise callers should be told when they are speaking to an AI, and said callers who stay engaged for a few turns gain confidence in the agent. The article cites no benchmarks or adoption figures.

Why it matters for digital humans: This is vendor commentary rather than data, but it is a useful read on where the bottleneck sits. Avatar products are downstream of the voice stack, so if latency and recognition accuracy are still the binding constraints, improvements in visual fidelity buy less than they appear to — a photorealistic face that mishears a name is worse than a plain voice agent, because the face raises the expectation it then fails. Gay's point about conveyed emotion is the uncomfortable one for the category: embodiment that adds no affective signal adds cost and uncanniness without adding function.