India, May 7 -- Krisp has launched Krisp VIVA 2.0, a voice AI infrastructure layer designed for voice agents, IVRs and conversational AI systems. The release introduces a new generation of small, real-time models aimed at improving word error rates (WER), predicting when users have finished speaking, classifying interruptions and detecting perceptual signals such as synthetic speech, gender and accent.

Voice agent adoption grew ninefold in 2025, yet many systems continue to struggle when deployed in real-world environments. Background noise and overlapping conversations can increase speech-to-text word error rates from around 5% to more than 30%. Voice activity detection systems may misinterpret background voices, fail to recognise genui...