Voice AI hasn't had its ChatGPT moment yet, executives say

Investors have poured billions of dollars into voice AI startups spanning model makers, enterprise customer-service tools, meeting note-takers and AI-powered dictation. New models and tools claiming to sound human arrive weekly. But according to two executives who spoke on stage at the HumanX conference last month, the technology still hasn’t reached the kind of breakout milestone that ChatGPT represented for text-based AI.

Shawn Wen, chief technology officer at enterprise voice AI platform PolyAI, said the industry has already cleared one hurdle: full-duplex models that can speak while listening to a caller. The next challenge, he said, is making the underlying reasoning fast enough that conversations feel natural. “We have reached the milestone of developing full-duplex models. The next challenge is to make reasoning very fast, so that the models can fetch answers quickly and the conversation feels natural,” Wen said. He added that customer-service agents built on these models need to sound confident rather than robotic — Wen said once a caller engages with an agent for its first two or three conversational turns, that caller tends to build enough trust to stop expecting a human on the line.

Alex Gay, chief marketing officer at meeting note-taker Otter, pointed to a different set of gaps: speaker identification, capturing intent, and tying that output to organizational knowledge. Otter is also working on digital twins that might represent people in meetings, which Gay said requires the output voice to carry the same emotive range as a real human — otherwise, he said, an avatar in a meeting becomes little more than a question-and-answer chatbot, incapable of the debate and strategic discussion that mark the best meetings.

Both executives said speech-recognition accuracy remains a weak link. Wen said automatic speech recognition (ASR) systems often miss important keywords, which breaks the broader context of a conversation. Gay said Otter keeps refining its own ASR model for the same reason: inaccurate transcription corrupts every downstream action a user takes based on it, eroding trust in the platform.

On transparency, both executives said voice tools should disclose when a conversation is being recorded or handled by AI. Otter said it wants to notify everyone in a chat when a meeting is being recorded, even if its own bot isn’t present, and PolyAI said establishing that enterprise callers are talking to AI is a priority.

Why it matters

The gap these executives describe — between heavy investor enthusiasm and voice AI’s unresolved reasoning speed and transcription accuracy — suggests enterprise buyers should expect further iteration before voice interfaces can reliably replace human agents.