TLDR: Krisp, a leader in real-time Voice AI technology, has launched VIVA, a new voice isolation AI model and software development kit (SDK) designed to significantly enhance the accuracy and naturalness of voice AI agent interactions. VIVA aims to improve ‘turn-taking’ behavior by eliminating background noise and secondary voices, a critical factor in seamless AI conversations, and is already processing over 1 billion minutes of voice audio monthly.
BERKELEY, Calif. – Krisp Technologies Inc., a prominent innovator in real-time Voice AI solutions, has officially announced the launch of VIVA, its groundbreaking voice isolation AI model and software development kit (SDK). This new offering is specifically engineered to elevate the precision and conversational fluidity of voice AI agents, addressing a critical challenge in human-AI interaction.
VIVA, an acronym for ‘Voice Isolation for Voice Agents,’ integrates directly into an application’s audio path. Its core function is to boost voice AI agents’ ability to accurately detect voice activity and refine their ‘turn-taking’ behavior. This capability is crucial for preventing false interruptions and fostering more natural, human-like conversations, a growing expectation among consumers interacting with AI systems. The company also revealed a significant milestone, stating that VIVA is currently processing more than 1 billion minutes of voice audio globally each month, underscoring its rapid adoption and impact.
According to Davit Baghdasaryan, co-founder and Chief Executive of Krisp, the challenge of ‘turn-taking’ is often severely compromised by background noise, particularly other voices. “Turn-taking gets really messed up when there’s background noise — especially background voices. The AI gets very confused,” Baghdasaryan explained in an exclusive interview. He elaborated that turn-taking, the natural back-and-forth flow of conversation, is instinctively managed by humans through verbal and nonverbal cues. For voice AI, it involves precisely detecting when a user has finished speaking to respond appropriately without interruption or awkward silence. Poor turn-taking can lead to frustrating and unnatural interactions.
Krisp’s VIVA model is designed to mitigate these issues by processing audio in under 20 milliseconds, dramatically enhancing responsiveness. This rapid processing contributes to an improvement in turn-taking accuracy by up to 3.5 times and is linked to a 50% reduction in dropped calls. The VIVA family of models is specifically tailored for server-side audio streams, integrating seamlessly in front of Voice Activity Detection (VAD) or Speech-to-Text (STT) systems. It effectively removes background noises and secondary voices, ensuring improved ‘turn detection’ and minimizing unwanted interruptions for Voice AI Agents and Bots, regardless of language or accent.
Also Read:
- Vonage and AWS Unveil Advanced AI Voice Agent Integration for Enhanced Business Communications
- Clarifai Advances Agentic AI with New Server Hosting and OpenAI Integration
This launch reinforces Krisp’s leadership in real-time voice isolation and noise cancellation, further solidifying its position in powering advanced voice AI products across the market. The VIVA SDK is part of Krisp’s broader suite of AI-powered voice technologies, which leverage deep neural networks to deliver unparalleled voice quality enhancements, transforming how businesses and communications are conducted.
