Nigerian-UK startup launches AI tool translating sign language to speech and text

Talksign, a Nigeria and UK-based artificial intelligence (AI) research and product company, has launched a sign language translation model, which the company claims converts American Sign Language into speech and text in under 100 milliseconds.

The company announced Talksign-1 on Monday, marking its first foundation model for sign language understanding. The model recognises 250 ASL signs and works bidirectionally, translating sign language to speech through webcam input and converting spoken or typed words into sign language video sequences.

The launch signals growth in more developers addressing a significant communication gap affecting millions globally. According to the World Health Organisation (WHO), over 430 million people worldwide are deaf, and 70 million people use sign language as their primary communication method, yet most digital tools, from video conferencing to customer service kiosks, still assume everyone can speak and hear, limiting access for the deaf and hard-of-hearing community in terms of getting a job, schooling and participating in the society. 

The assistive technology could enable direct communication between deaf and hearing individuals in everyday settings without waiting for human interpreters, joining startups like SignVrse, which are using artificial intelligence to reduce the barrier between persons with hearing impairment. 

Edidiong Ekong and AI engineer Kazi Mahathir Rahman founded Talksign in November 2025 to address the communication barriers faced by the deaf and hard-of-hearing community in accessing digital tools and everyday services. The company’s model works bidirectionally, translating sign language to speech through webcam input and converting spoken or typed words into sign language video sequences.

The system works by capturing hand, body and face movements through a standard webcam. Talksign-1 analyses about one second of signing before making predictions, balancing speed with accuracy. The model was trained on the WLASL2000 dataset, a large-scale collection of ASL signs.

In testing, the company said the model achieved 84.7% accuracy on single signs. However, the company acknowledges that the system does not yet support continuous sentence-level interpretation or fingerspelling, which limits its use to isolated signs rather than full conversations.

“We believe AI should augment, not replace human interpreters,” Ekong said. 

Ekong said his interest in sign language accessibility came from his childhood experiences in Nigeria.

“I grew up with three deaf friends in Nigeria,” he said. “At age 9, I realised I was being left out of their world, not the other way around. So I learned American Sign Language to join their conversations. By age 12, I was teaching sign language to hearing people, running classes for 10 years.”

The company identified potential applications in education, healthcare, workplace settings and public spaces, including transport announcements, emergency alerts, and live news broadcasts.

“We believe accessibility is a human right, not a feature,” Ekong said.

The company said it partnered with deaf educators, native ASL signers, and accessibility advocates throughout the development process. Landmark extraction happens on the user’s device in the browser, meaning only processed data points, not raw video, are sent to company servers for analysis, according to the company’s privacy policy.

The company cautioned that the technology “should not be used as a sole authority in medical, legal, or safety-critical settings without human oversight.”

Talksign-1 currently supports only ASL, though the company said it is developing support for British Sign Language, French Sign Language and other languages. It is also working on expanding the vocabulary beyond 250 signs and developing capabilities for recognising continuous signing and fingerspelling in future versions.