Talksign, an AI company operating across Nigeria and the UK, has shipped two models that translate between American Sign Language and text or speech in real time. Palm 1.0 and Echo 1.0 launched on May 20. One reads sign language. The other produces it.
The WHO puts the number of people with disabling hearing loss at over 430 million globally. Tens of millions use sign language as their primary mode of communication. Most digital tools still assume users can hear and speak. These models are built for the gap that assumption creates.
READ: Nigerian Startup Smartcomply Enters London with African AML Tech
Palm 1.0 Reads ASL With 84.2% Semantic Accuracy
Palm 1.0 translates American Sign Language into text or speech in real time. Talksign reports 84.2% semantic accuracy. The model captures what a signer means, not just what gesture appears on screen. Word-level accuracy sits at 79.6%.
Those numbers matter because ASL has its own grammar, its own syntax, and a historical shortage of large-scale training data. Most translation tools break down at the sentence level. Palm 1.0 was built specifically to handle continuous signing.
Palm 1.0 trained on over 71,000 ASL samples, using a transformer-based architecture built around SAGE, Spatial Attention Graph Encoder. SAGE maps 133 anatomical landmarks across the hands, head, and shoulders. The system reads signing as continuous movement, not isolated gestures.
"Palm 1.0 is the first model we are confident putting into the hands of Deaf users at scale," said CEO and co-founder Edidiong Ekong. The company plans to deploy it on phones, smart glasses, in classrooms, and in hospitals.
Echo 1.0 Generates Photorealistic Signing at 29 Milliseconds
Echo 1.0 takes English text or speech and outputs a photorealistic ASL video at 30 frames per second. Translation latency runs at approximately 29 milliseconds. The result is that it appears in real time.
The model was trained on 94,410 ASL sentence pairs, running through the full dataset 15 times. Echo 1.0 does not transliterate English word by word. It converts text into ASL gloss first, preserving sign language grammar and structure, then maps each gloss token to a 3D motion sequence rendered by a neural engine.
Echo 1.0 also generates personalised avatars from a single reference photo. The feature lets Deaf users interact with a face they recognise rather than a generic avatar. CTO Kazi Mahathir Rahman said the models open the door to sign language becoming a first-class interface for AI systems. That removes the assumption baked into most AI systems that users must type or speak.
Landmark extraction happens on the user's device. Only processed data points reach company servers; raw video does not leave the phone.
READ: OpenAI Offers YC Startups $2M in Tokens - Equity Required
What Talksign Still Needs to Solve
Talksign has not papered over what the models cannot yet do.
Echo 1.0 accepts English only for now. Spanish, French, and Arabic support are planned. Specialised vocabulary in medicine, law, and engineering requires separate fine-tuning. Multi-word ASL phrases remain only partially modelled. Palm 1.0 handles sentence-level signing but is not yet optimised for every continuous signing context.
The company's earlier model, Talksign-1, launched in February and covered 250 ASL signs. It could not handle continuous sentences or fingerspelling. Palm 1.0 and Echo 1.0 were built to fix exactly that.
Nigerian Sign Language is already on the roadmap alongside British Sign Language and German Sign Language for future versions. Full deployment on desktop and Meta Ray-Ban smart glasses is scheduled for August 20, 2026.
READ: Konza Signs North Africa Deal as Kenya Passes Tech City Law
What This Means for Nigeria
Nigeria has one of the largest populations of people with hearing impairments on the continent. Professional sign language interpreters remain scarce outside major cities. A system that works on a phone in real time without an interpreter present changes the practical calculus for education, healthcare, and emergency communication across the country.
Talksign joins SignVrse and other AI accessibility platforms building in the same space. A Nigeria-founded team shipping production-grade models at these benchmarks is a different kind of signal that the African tech ecosystem should pay attention to.
