Paraspeak (India) | Real-time AI voice conversion and correction

Just 18 months ago, Pranet Khetan went on a high school field trip to the Indian Stroke and Paralysis Foundation (ISPF) in Sukhbir Nagar, Delhi. Little did he know that the experience would end up with him winning the honor of becoming a Solve for Tomorrow ambassador.

During the visit, Pranet met patients living with all types of conditions, but he identified a critical throughline: many were stripped of the fundamental human ability to communicate. After careful research, he realized that for the hundreds of millions of non-English speakers with impaired speech, there was simply no proper voice-assistive technology available. And so, the idea for Paraspeak was born: an AI-powered wearable that converts impaired speech in near real-time, giving a voice back to people with dysarthria and other voice and speech disorders, especially those who speak underserved languages. As a "voice-first" wearable, it doesn’t require them to type or select icons, which is often physically exhausting. Instead, Paraspeak just listens.

Turning Impaired Speech Into Clear Communication

Imagine being trapped inside your own body. Your mind is sharp, your thoughts are complex, but you cannot ask for water, express pain to a doctor, or communicate your thoughts to the people around you. This inability to communicate can lead to withdrawal into oneself, a lack of education, and a whole range of poor healthcare outcomes. What’s more, the scale of this issue is staggering, with over 650 million people worldwide suffering from voice and speech disorders.

Paraspeak offers a real-time, speaker-independent speech recognition and enhancement system for individuals with dysarthria and other conditions. It records their slurred speech, runs it through a state-of-the-art deep learning model, and plays out clear speech for everyone to hear, all in a small, low-cost, IoT-enabled device. The product allows people to communicate effectively with those in their surroundings, with a special emphasis on under-researched languages and low-resource settings.

Beyond the immediate utility of the device, Paraspeak aims to fundamentally change the stigma around disability. When a person uses it and speaks clearly, the listener realizes: “They never lacked intelligence, they only lacked an interface.” This realization shatters the bias that links speech impairment with cognitive decline, and it’s why the project’s slogan is “Every. Voice. Deserves. Clarity.”

How Samsung Solve for Tomorrow Accelerated Paraspeak’s Development

The journey of Paraspeak has been a continuous process of refining a raw idea into a scalable medical system. It began with a fundamental gap, the lack of data, so Pranet’s first milestone was building a custom data collection toolkit to compile India’s first database of Hindi dysarthric speech, 1,400+ samples from 28 patients. Then, he took the project forward from theoretical experiments to a functional, patent-pending deep learning framework.

The technical challenges were immense, with early versions of the models showing error rates greater than 4,000%, so Pranet built custom intelligibility analysis and data augmentation pipelines to achieve a word error rate of only 3.3%. Then, when he realized that most patients would be unable to use a model deployed as an app due to other symptoms, he decided to create a small, low-cost IoT device instead.

During his Solve for Tomorrow participation, Pranet brought improvements to the system to make it more market-ready, advancing the technology and building out the core software, with the mentorship and prize money acting as key platforms for his progress. In particular, a significant shift from research to product development happened as a result of the program. Guidance from industry experts at Samsung pushed him to look at design and human-computer interaction, resulting in the reiteration of two major hardware prototypes. He dove deep into industrial design, transitioning from a fragile lab prototype to a more robust, wearable form factor, discussing critical real-world features like waterproofing and battery efficiency.

Pranet also realized his model wasn't just for dysarthria, which is the disorder he was initially targeting. He was able to expand the core capabilities to cover Presbyphonia, where the voice degrades with age. This significantly widened the potential impact, with plans to cover even more voice and speech disorders.

Building a More Accessible World

With Paraspeak already demonstrated in real-world patient settings, the next phase focuses on clinical trials and large-scale data collection in 2026, followed by regulatory approval and an initial market launch in 2027. Beyond the device itself, Paraspeak aims to scale globally, expanding languages, entering hospitals and public institutions, and opening its API so impaired speech can be understood everywhere, not just through one product.

When it comes to other young innovators, Pranet’s advice is straight to the point: “Start small, stay curious, and keep dedicated to solving real-world problems. Progress, no matter how small, will add up.” After all, he has work to do.

Samsung Solve for Tomorrow Ambassadors will be officially appointed at a ceremony held in Milan on Feb. 10-11, in conjunction with the International Olympic Committee (IOC) Find out more about Solve for Tomorrow at [link].