Voice AI hasn't had its "ChatGPT moment" yet, industry executives say
Executives at PolyAI and Otter say voice AI has yet to reach its "ChatGPT moment" despite the arrival of full-duplex models. Fast reasoning, accurate speech recognition and transparency remain the key challenges.
Investors have poured billions of dollars into voice artificial intelligence (AI) startups, from model developers and customer service platforms to meeting note-takers and dictation tools. Yet two industry executives argue the technology has not reached the breakthrough that ChatGPT delivered for text-based AI.
Speaking at the HumanX conference last month, Shawn Wen, chief technology officer of PolyAI, a voice AI platform for enterprise customer service, said the industry has reached the milestone of building full-duplex models, which can speak while listening to the other party. In his view, the next challenge is making reasoning very fast, so models can retrieve answers quickly and conversations feel natural.
Building trust in customer service
Wen stressed that customer service AI agents should not sound robotic and should give callers confidence that their problem can be solved. He said that once the voice is good enough and a customer is willing to engage for the first two or three turns, confidence starts to build, and over time people feel they may not need to speak to a human.
Alex Gay, chief marketing officer of meeting note-taking app Otter, said identifying speakers, capturing intent and linking it with organizational knowledge is a key step toward automation. The company is also working on digital twins that could represent people in meetings. For that, Gay said, the synthetic voice must convey the same emotional expression as a human; otherwise it is just a question-and-answer chatbot.
Recognition accuracy and transparency
Wen believes automatic speech recognition (ASR) models often miss important keywords, which prevents them from capturing the full context. Gay agreed, saying Otter keeps improving its transcription because errors make all follow-up actions flawed and erode trust in the platform. He also named language as an area where voice models need to improve.
Both executives also addressed transparency: customers should be told when they are being recorded or are talking to an AI. Otter wants to extend this to meetings where its bot is not present, for example by notifying participants in the chat that a recording is under way.
