Suvana: An Integrated Platform for Two-Way Deaf–Hearing Communication in Sri Lankan Sign Language

In many parts of the world, communication is a shared, fluid exchange that happens effortlessly between people who can hear and those who cannot. But for the Deaf community in Sri Lanka, this exchange is often blocked by a lack of resources. There are not enough professional interpreters to bridge the gap, and the country does not yet have a single, fully standardized version of its sign language that everyone agrees upon. While technology has made great strides in helping people communicate, most of these tools are built for American Sign Language, leaving local languages like Sri Lankan Sign Language largely ignored by developers. This creates a situation where the people who need these tools the most are left without them, and the technology that does exist often fails to understand the unique ways people in Sri Lanka express themselves with their hands and faces.

A researcher has set out to address this gap with a new digital platform called Suvana. Rather than trying to build a single, massive application from scratch, they took four different software tools that had been developed separately and stitched them together into one seamless experience. The goal was not to claim they had solved the problem of communication barriers entirely, but to demonstrate that these distinct pieces could work together as a single product. The platform includes a tool that turns hand signs into spoken Sinhala, a feature that converts spoken Sinhala into a digital avatar signing back, a mobile app that alerts users to important sounds like car horns or breaking glass, and a practice tutor for hearing people who want to learn the language. What makes this project notable is not just the combination of these tools, but the honesty with which the researcher reports what they have achieved and, perhaps more importantly, what they have not yet proven.

The most significant finding of the work comes from the practice tutor, a tool designed to help learners practice signs and receive feedback on their performance. The researcher built a system that compares a learner's hand movements against a reference video of a fluent signer. To do this, they used a method that aligns the two sequences of movement in time, allowing for differences in speed so that a slow but correct sign is not marked down for being slow. However, during their testing, they discovered a critical limitation in how this technology works. They found that some signs that are completely different from each other can look surprisingly similar to the computer, while two people performing the exact same sign correctly can look quite different from one another.

Because of this, the researcher realized that the system cannot safely be used to guess what sign a person is trying to make if the computer does not already know what they are supposed to be signing. If the system tried to identify an unknown sign, it would likely make mistakes because the distance between two different signs is sometimes smaller than the distance between two correct attempts at the same sign. Instead, the system is designed to work differently: the learner is asked to practice a specific sign, and the computer simply grades how well they performed that specific sign. This approach avoids the trap of guessing the wrong word. The researcher measured how well the system could tell the difference between a correct attempt and a wrong one, finding that it could distinguish between them with a level of accuracy that was significantly better than random chance, though still far from perfect.

The researcher was also very careful about what they claimed regarding the other parts of the platform. For the tool that turns signs into speech, they noted that the system works well when the person signing is the same one the computer was trained on, but they did not claim it would work for anyone else. They explicitly stated that they cannot yet say if the system would work for a new person they have never seen before, because the data they have is limited. Similarly, for the part of the system that turns speech into a signing avatar, they were clear that the digital character's movements were created by a hearing student and have not been checked by a Deaf signer. They present this avatar as a demonstration of how the technology works, not as a validated way to communicate, acknowledging that without input from the Deaf community, they cannot say if the signs are actually understandable.

Perhaps the most important part of the paper is what the author says about their own role in the project. All four researchers are hearing people, meaning they do not have the lived experience of being Deaf. They openly admit that they lack the fluency and the cultural context to judge whether the system they built is truly useful or acceptable to the people it is meant to help. They point out that in the field of accessibility technology, it is common for hearing researchers to build tools without enough input from the Deaf community, a practice that can lead to products that miss the mark. They describe their current work as a step toward a solution, but they emphasize that the next steps must involve Deaf signers directly. This means recording new reference videos with Deaf teachers and having them validate the feedback the system gives.

The researcher concludes that while they have successfully integrated these four tools into one platform and calibrated the scoring system based on real data, the work is far from finished. They have measured how the system performs against a set of recorded videos, but they have not yet measured how it performs with real learners or how well it helps people learn. They have not tested the system with a diverse group of signers, nor have they validated the content with the Deaf community. Their findings serve as a clear map of what is possible with current technology and what remains unknown. The path forward, they argue, requires moving beyond just building the technology to ensuring that the people who will use it are the ones guiding its development. Until that happens, the system remains a promising prototype rather than a complete solution, and the researcher is clear that the true test of their work will only come when Deaf signers are involved in every stage of the process.