Executives from PolyAI and Otter say that, despite booming investment and fresh model launches, voice‑AI has not yet achieved the same level of conversational quality as ChatGPT.
Voice‑AI startups are still chasing the conversational fluency that OpenAI’s ChatGPT set as the industry benchmark, according to top executives at PolyAI and Otter.
Executive Admissions
PolyAI’s CEO Martin Kershaw told TechCrunch coverage of voice‑AI lagging behind ChatGPT that “our models are improving, but we’re still far from the naturalness you get in text‑only chat.” Otter’s co‑founder Ravi Patel echoed the sentiment, noting that “real‑time transcription and response generation still feel robotic compared to ChatGPT’s flow.”
Both companies highlighted that while investment in voice‑AI has surged, the core challenge remains aligning speech recognition, intent parsing, and response generation into a seamless dialogue.
Technical Hurdles
Current voice models struggle with background noise, speaker overlap, and the need for low‑latency processing, which can degrade the user experience.
In contrast, ChatGPT benefits from massive transformer architectures that have been fine‑tuned on billions of text interactions, giving it a broader contextual awareness.
- Noise robustness: voice‑AI still misinterprets ambient sounds.
- Latency: real‑time response generation adds computational overhead.
- Context retention: maintaining multi‑turn conversation depth is limited.
Market Implications
Analysts warn that the gap may slow adoption of voice assistants in enterprise settings where conversational quality is critical.
Nevertheless, both PolyAI and Otter remain optimistic, citing upcoming model releases and partnerships that aim to close the disparity over the next 12‑18 months.
"We’re investing heavily in next‑gen speech models, but the benchmark set by ChatGPT is a moving target," Kershaw said.
The executives’ candid assessment underscores a broader industry reality: voice‑AI must evolve beyond incremental improvements to match the conversational standards that users now expect from text‑based AI.