Krisp's AI Converts Accents in Real-Time for Clearer Communication

03/26/2025 Artificial Intelligence

Krisp, an audio innovation company, has just unveiled a fascinating new feature: AI-powered accent conversion. Imagine speaking in your native accent and having it seamlessly transformed into another in real-time during calls. That's precisely what Krisp aims to achieve, starting with Indian English accents converted to U.S. English.

According to Krisp, this isn't just a superficial change; the AI carefully swaps out phonemes to align with American accents while preserving the speaker's unique voice. This technology has reportedly undergone testing in enterprise settings, and now, a beta version is accessible through the Krisp desktop application. Users have the flexibility to activate the feature whenever they need, either before or during a call.

The Origin of the Accent Converter

Arto Minasyan, co-founder of Krisp, shared the personal motivation behind this development. He noted that his accent sometimes posed a barrier to clear communication, even when speaking English fluently. This inspired the team to explore accent conversion as a potential solution, leading to two years of dedicated work and the beta release we see today.

While initial testing by some reporters yielded mixed results – with occasional unnatural sounds and missed words – Krisp attributes these issues to the beta stage and assures users that the AI model will continuously improve with more data and user feedback.

Promising Results in Enterprise Testing

Krisp highlights impressive results from enterprise testing, claiming a 26.1% increase in sales conversion rates and a 14.8% boost in revenue per booking. The decision to focus on Indian accents initially stems from the significant representation of Indian professionals in STEM fields globally. The company also plans to expand support to include other accents, such as Filipino.

Other companies, such as GV-backed Sanas, have also been developing similar technology, deploying it at scale in call centers. Krisp trained its model using vast datasets of speech samples encompassing diverse accents and dialects, incorporating data from its meeting assistant with user consent.

A key advantage of Krisp's approach, according to Minasyan, is that it doesn't require pre-training on individual voices. The AI creates a speaker profile in real time, making it instantly usable.

Looking ahead, Krisp, which last secured funding in 2021, intends to launch iOS and Android applications this year to facilitate in-person meetings. Furthermore, a new Chrome extension is in development to enhance integration with Google Meet.