Voice AI company Krisp has launched VIVA 2.0, a new version of its voice infrastructure platform designed to improve the reliability of AI-powered voice agents and conversational systems operating in real-world environments.
The company said the updated platform introduces a new generation of lightweight, real-time audio models that help AI systems better understand interruptions, detect accents, identify synthetic speech, and predict when users have finished speaking.
The launch comes as enterprises increasingly adopt voice AI systems for customer service, IVRs and automated support operations. According to the company, voice agent usage grew 9x in 2025, but many systems continue to struggle outside controlled demo environments due to background noise, interruptions, and speech recognition errors.
“Voice is becoming the primary interface between humans and AI,” said Robert Schoenfield, EVP of Licensing and Partnerships at Krisp. He added that those conversations don’t happen in clean environments. They happen in the real world, shaped by noise and subtle human cues.
Krisp said VIVA 2.0 works as an infrastructure layer, sitting between speech-to-text and language models, helping clean and interpret audio signals before AI systems process them.
One of the key additions is Turn Prediction v3, a multilingual model that predicts when a speaker has completed a sentence without relying on text transcription. The company said this helps reduce awkward interruptions and improves conversational flow.
The platform also introduced Interrupt Prediction, which distinguishes between genuine interruptions and simple conversational acknowledgements such as “hmm” or “yes.”
Krisp has also added a set of real-time “signal detectors” capable of identifying accents, speaker gender and AI-generated speech. According to the company, the accent detection feature can route audio to speech recognition systems optimised for specific accents, improving transcription accuracy.
The upgraded Voice Isolation v3 model aims to reduce background noise and improve speech recognition quality during calls.
The company noted that the platform currently processes over 12 billion minutes of voice AI traffic annually and is integrated into more than 130 voice AI products, including Telnyx, LiveKit, and Vapi.
Your reaction
Discussion
What's your take on this story?
No comments yet — be the first to share your view.