Executives say voice AI still struggles with understanding and trust

In short: Some voice AI leaders say today’s voice assistants still make basic mistakes and have not yet reached a widely trusted, natural experience.

What's going on

Investors are putting billions of dollars into voice AI, which is software that talks and listens like a person. New tools often claim they can hold natural conversations. But executives at PolyAI and Otter said the reality is still messy.

Shawn Wen, CTO of customer service voice AI company PolyAI, said even with newer systems that can talk while listening, the next problem is speed. He said the system has to “think” fast enough that replies feel natural, like a real conversation instead of awkward pauses.

Both Wen and Alex Gay, CMO of meeting note taker Otter, pointed to a core issue: the software often misunderstands what people say. Wen said ASR, short for Automatic Speech Recognition, can miss important words. That is the speech to text part (like a very fast typist listening to you). If it gets the words wrong, everything that follows, such as summaries or next steps, can also be wrong.

They also raised trust and transparency concerns. Otter said it wants clearer ways to notify everyone in a meeting when recording is happening. Wen said companies should clearly tell customers when they are talking to an AI on support calls.

What to watch

Watch for improvements in transcription accuracy, faster response times, and clearer notices that AI is present. If those pieces improve, more people may accept AI for customer service calls and meeting follow-ups.

Source: TechCrunch AI