Krisp AI: Sound American on Calls - Voice Changer for Indians

Krisp Introduces AI-Powered Accent Conversion

Krisp, an audio technology company, announced on Wednesday the launch of a novel feature leveraging AI to modify a user’s accent during voice calls.

Initially, the functionality will focus on converting Indian English accents to a U.S. English pronunciation.

Preserving Voice Identity

The company asserts that the accent conversion process is designed to retain the speaker’s unique vocal characteristics.

It achieves this by altering only the phonemes to align with American English pronunciation patterns.

This feature has undergone testing within enterprise settings, and a beta version is now available to users of the Krisp desktop application.

Users have the flexibility to activate the feature either before or during calls.

The Genesis of the Feature

According to Arto Minasyan, Krisp’s co-founder, the concept originated from personal experience.

“I frequently encountered situations where my accent hindered clear communication, despite proficient English speaking skills,” Minasyan explained.

“We hypothesized that accent modification could significantly improve mutual understanding.

Development commenced two years ago, culminating in this beta release.”

Initial Beta Testing Results

Early testing by a reporter revealed that the processed voice occasionally sounded unnatural and experienced instances of missed words.

Krisp acknowledged these shortcomings, attributing them to the beta nature of the release and promising ongoing model improvements.

Reported Enterprise Benefits

Krisp reports positive outcomes from enterprise trials of the accent conversion feature.

Specifically, they observed a 26.1% increase in sales conversion rates and a 14.8% rise in revenue per booking.

Focus on Indian Accents and Future Expansion

The company prioritized Indian accents due to the substantial representation of individuals from India within the global STEM workforce.

Plans are in place to extend support to additional accents, with Filipino accents being among the next to be implemented.

Similar technologies are already being utilized by other startups, such as Sanas, within large-scale call center operations.

Model Training and Real-Time Adaptation

The AI model was trained using a vast dataset of speech samples encompassing diverse accents and dialects.

Data utilized for training was sourced from Krisp’s meeting assistant, with explicit user consent obtained.

A key advantage of this feature is its ability to create a speaker profile in real-time, eliminating the need for pre-training on individual voices.

Future Developments

Krisp, having last secured funding in 2021, intends to launch iOS and Android applications this year to facilitate support for in-person meetings.

Furthermore, a new Chrome extension is under development to enhance integration with Google Meet.