Science & Tech · AI · 15 hrs ago
Voice AI still needs faster reasoning and more reliable understanding
Voice AI tools can now speak while listening, but executives at PolyAI and Otter say the technology is not yet reliable enough to feel like natural conversation.
PolyAI develops voice systems for businesses, including customer service, while Otter makes meeting note-taking tools.
PolyAI CTO Shawn Wen says models need to reason faster so they can answer quickly and keep conversations flowing.
Otter’s Alex Gay says accurate transcripts and understanding who said what are important for turning meeting conversations into useful actions.
Mistakes in speech recognition can lead to flawed summaries or actions and reduce users’ trust.
The executives also say AI tools should make clear when people are speaking to AI or being recorded.
They expect voice AI to become more useful as its speed, accuracy and ability to build confidence improve.
Voice AI executives say the technology has not yet reached a breakthrough comparable to ChatGPT.
PolyAI CTO Shawn Wen says the next challenge is making models reason quickly enough for natural conversation.
Executives also point to problems with speech recognition, language understanding and transcription accuracy.
Otter CMO Alex Gay says inaccurate transcripts can undermine follow-up actions and trust in the platform.
The article says voice tools should be transparent about recording people or using AI.
- Who
- PolyAI CTO Shawn Wen and Otter CMO Alex Gay.
- What
- They say voice AI still needs faster reasoning and more reliable understanding to support natural, trusted interactions.
- When
- The article was published on 2026-10-11; Wen spoke at the HumanX conference the previous month.
- Where
- The comments were made on stage at the HumanX conference.
- Why
- Slow reasoning and errors in speech recognition and transcription can make conversations feel unnatural and lead to flawed actions.
This story does not have two clearly opposing sides.
The next challenge is to make reasoning very fast, so that the models can fetch answers quickly and the conversation feels natural,
But if your original transcription didn’t have the accuracy that you needed, all the follow-up actions that you have become flawed.
You lose trust in the platform.
Shawn Wen spoke at the HumanX conference about the challenges facing voice AI.
TechCrunch published the article about executives’ views on voice AI.
- PolyAI CTO
- Shawn Wen
- Otter CMO
- Alex Gay
- Conference
- HumanX
- Publication date
- 2026-10-11
- Speech recognition
- Automatic Speech Recognition (ASR) models can miss important keywords
Sources
Related news
Anthropic AI submits false murder tip to Philadelphia police
Tech leaders adopt Trump’s preferred term “super intelligence”
AWS releases open-source sandbox for AI agents in developer preview
Nadella calls for human-controlled emergency brake on advanced AI
TikTok CEO calls for stronger safeguards against AI cybersecurity risks
European AI startups seek ways to compete with US and China
US verdict re-examined over AI-generated video of victim
Three-quarters of online stores score poorly on AI shopping readiness
TikTok CEO predicts AI-driven scientific breakthroughs within three years
Microsoft expands Copilot Notebooks with productivity and automation upgrades
Chinese factories are adopting AI agents and taking the approach abroad
