Krisp's New AI Delivers Auditable Accuracy in Voice Translation

BERKELEY, CA – June 09, 2026 – In a world increasingly connected by technology but divided by language, the promise of seamless real-time communication has often felt just out of reach. Today, Berkeley-based Krisp, a leader in Voice AI, has taken a significant step toward closing that gap, not just with a technological leap, but with a strategic one. The company announced the launch of Voice Translation v3 for enterprises alongside a self-serve Voice Translation API, effectively making its high-stakes, battle-tested translation engine accessible to everyone from global contact centers to solo developers.

The core of the announcement is an engine that has already proven its mettle in one of the most demanding environments imaginable: a live healthcare deployment. There, it achieved a 96% overall translation accuracy rate, handling 90% of multilingual calls end-to-end without a human interpreter and, most critically, without a single patient safety incident. Now, that same engine is being offered with a new layer of operational control for businesses and a simple on-ramp for developers, shifting voice translation from a novel feature to a core, auditable infrastructure.

The Accuracy Imperative in a Multilingual World

Real-time voice translation is a notoriously difficult technical challenge. The clean audio of a lab demonstration rarely reflects the reality of a call from a busy street, a hospital room, or a noisy call center. Accents, domain-specific jargon, and background noise have long been the Achilles' heel of automated systems.

"Real-time voice translation is having its moment, but most of what is shipping was built on general data and not tested where accuracy matters," said Davit Baghdasaryan, CEO and Co-Founder of Krisp. "We built our engine for the most difficult environments: live calls in healthcare, insurance, and financial services where one wrong word has real consequences."

This focus on "difficult environments" is what makes the 96% accuracy figure so compelling. It’s a number not born from pristine data sets, but forged in the chaos of real-world application. When compared to industry benchmarks—where even top-tier systems from tech giants like Google hover around 94% for specific language pairs like Spanish medical instructions, and general-purpose tools can dip to 65-80% in noisy conditions—Krisp’s claim stands out. The company reports that its internal benchmarks, independently confirmed by bilingual linguists, show consistent 93-97% accuracy across 30 languages and six business domains. This isn't just about getting the gist of a conversation; it's about preserving the precise meaning of every critical word.

From Feature to Infrastructure: A New Level of Control for Enterprises

For the enterprises and Business Process Outsourcing (BPO) providers that Krisp’s Voice Translation v3 targets, the challenge has never been just about accuracy. It’s about accountability. How can a business prove that its translations are compliant? How can a manager audit a multilingual call for quality? How can regulated content be delivered flawlessly in any language?

Voice Translation v3 directly addresses these questions with a suite of operational controls that transform translation into a measurable and manageable part of business operations. The new "Accuracy QA" feature automatically scores 100% of translated calls, providing a level of oversight previously impossible at scale. This allows operations leaders to move beyond spot-checks and gain comprehensive insight into quality across their entire organization.

In regulated industries like healthcare and finance, this is a game-changer. Features like "Quick Phrases" allow pre-written, legally-vetted content—such as medical disclaimers or financial disclosures—to be delivered as perfectly translated speech, eliminating the risk of human or machine error on critical information. The "Live Call Audit" function gives administrators a real-time window into conversations, with access to a live bilingual transcript, enabling immediate intervention or post-call review for training and compliance. Furthermore, the ability to build a "Custom Vocabulary" ensures that industry-specific terms, from medical diagnoses to financial instruments, are recognized and translated with precision.

This shift moves the technology beyond a simple communication aid. For a hospital, it’s a tool for patient safety and equitable access. For a bank, it’s a mechanism for compliance and risk mitigation. For the BPOs that serve them, it’s a powerful way to offer reliable, auditable multilingual support, expanding their service capabilities while managing risk.

Opening the Gates: Empowering Developers with Battle-Tested AI

While the enterprise solution locks down control and compliance, the parallel launch of the Voice Translation API does the opposite: it opens the floodgates for innovation. By offering its core engine through a self-serve API, Krisp is giving developers direct access to technology that was previously the domain of large-scale enterprise deployments.

"Developers building telehealth, customer support, fintech, and other accuracy-critical products need more than a demo that works on clean audio," noted Robert Schoenfield, EVP of Licensing and Partnerships at Krisp. "They need an engine that has already been tested in live, high-stakes calls. The Voice Translation API gives them direct access to that engine, with a developer experience that does not require a sales call to get started."

The developer experience is designed for simplicity and speed. A single WebSocket handles audio input and delivers translated speech and text output. SDKs for JavaScript and Python are available at launch, with C++ on the way. Developers can sign up, get an API key, and start building immediately, with 60 free minutes to test the service.

This accessibility could fuel a new wave of applications where accuracy is non-negotiable. Imagine telehealth platforms that instantly and reliably connect doctors with patients regardless of their native tongue, or fintech apps that can provide secure customer support across the globe. The potential extends to global collaboration tools, educational technology, and even multiplayer gaming, where clear, real-time communication can create more inclusive communities. By providing a proven, high-accuracy engine as a simple building block, Krisp is betting that developers will find uses for it that no one has even thought of yet.

Underpinning Trust with Privacy and Security

In any discussion involving voice data, especially in regulated fields, the conversation inevitably turns to privacy and security. Krisp addresses this head-on with a "privacy-first" architecture that is fundamental to its design. By default, all voice data is processed exclusively in memory and is not stored by the company. This in-transit processing model is crucial for earning the trust of enterprises concerned with data sovereignty and confidentiality.

For organizations that require auditable records for quality assurance, Krisp offers an optional feature to store call transcripts. However, control remains firmly in the hands of the client, who can delete all stored data at any time. This entire system is buttressed by robust security measures, including end-to-end encryption and a suite of certifications including SOC 2 Type II, GDPR, and HIPAA compliance.

This deep-seated focus on security and privacy, combined with auditable accuracy and newfound accessibility, represents a maturation of voice AI technology. It's a move from the realm of possibility to the world of practical, reliable, and tangible difference, offering a clearer voice for a multilingual world.