Talksign wants to simplify sign language for deaf people with its AI solution

Modern artificial intelligence can draft complex code, generate photorealistic video, and translate spoken words in milliseconds. Yet, it fundamentally fails at understanding the natural communication of hundreds of millions of people.

The barrier is not compute power or algorithmic sophistication. The problem is a severe lack of structured, licensable data.

While leading speech models train on decades’ worth of scraped open-web audio, the infrastructure for multimodal sign language data has remained virtually nonexistent. Most publicly available sign language corpora are tiny, recorded in sterile studio environments, and limited to a handful of signers. When developers force these narrow datasets into machine learning models, the products inevitably fail in the real world. They struggle with regional dialects, native signing rhythms, and the complex facial grammar required to convey intensity or ask a question.

Talksign is building the missing infrastructure to solve this. 

Edidiong Ekong, Talksign founder and CEO, in an exclusive statement shared with Technext, says, “Today we’re launching the Talksign Marketplace, a place where Deaf communities, researchers, institutions, and companies can contribute, annotate, license, and access high-quality sign language datasets. It’s the missing infrastructure layer beneath every serious attempt to build AI that works for Deaf people, and it’s the piece we’ve had to build ourselves to build anything else.

Edidiong Ekong, co-founder and CEO of Talksign

For the global tech industry, and particularly the African artificial intelligence ecosystem, this launch represents more than just an accessibility tool. It is a commercial and ethical framework for how we solve the low-resource language problem in AI,” he adds.

The infrastructure for embodied language

According to World Health Organisation data, over 430 million people globally experience disabling hearing loss. These populations rely on between 200 and 300 documented sign languages. Despite this scale, sign languages are entirely unrepresented in the current machine learning boom.

Sign languages are full natural languages with unique morphology and syntax, expressed simultaneously across the hands, face, upper body, and spatial environment. Building AI that understands this requires solving a genuinely difficult multimodal computing problem. Models must reason across multiple visual channels in real time.

The data needed to train these models exists, but it is deeply siloed. It sits locked in university archives without commercial licensing pathways or remains in the hands of independent Deaf organisations that have never been offered a financial incentive to share it.

Talksign’s marketplace functions across four core pillars to commercialise and standardise this data:

• Contribute: Independent signers and Deaf organisations can submit recorded video data and receive direct financial compensation.
• Annotate: Raw video is heavily processed into training data through gloss annotation, pose extraction, and linguistic labelling. Talksign routes this paid work directly to fluent signers rather than outsourcing it to non-native workers.
• Licence: Buyers receive machine-readable licensing terms and clear provenance records, solving a massive compliance headache for enterprise AI builders.
• Access: Startups and researchers can evaluate and acquire diverse datasets across multiple dialects and recording environments from a single hub.

Talksign Marketplace: A blueprint for the African AI ecosystem

For developers and founders in Africa, the underlying problem Talksign is solving will sound intimately familiar. The continent is home to thousands of spoken and signed languages, including Nigerian Sign Language (NSL), which suffer from the same data scarcity, locking them out of the global AI economy.

African technologists building natural language processing tools for languages like Hausa, Yoruba, or Swahili consistently hit a data wall. The Talksign framework offers a compelling blueprint for how local founders can ethically aggregate and monetise low-resource language data. By creating a transparent marketplace that pays native speakers and signers for their contributions, developers can bypass the slow, grant-dependent routes of academic data gathering and build robust, commercial-grade corpora.

Talksign Marketplace

The historical relationship between the tech industry and marginalised communities is largely extractive. Data is scraped, models are trained, and the communities generating the raw material receive nothing.

Because sign language data is inherently biometric, capturing a person’s face and body, it cannot be anonymised like text. Talksign has introduced an open verification portal to address this directly. The system ties every dataset back to its source and licensing terms, ensuring auditable consent and likeness rights. Buyers can verify the fluency of the signers and the accuracy of the annotations before integrating the data into their pipelines.

Talksign claims this infrastructure was born out of its own internal development hurdles. The company points to its proprietary models as proof of concept. Palm 1.0, a speech-to-sign model, reportedly reaches 84.2% semantic accuracy at roughly 29 milliseconds of latency. Its sign-to-speech counterpart, Echo 1.0, reportedly hits a 0.927 Structural Similarity Index at 30 frames per second.

Ultimately, captioning is a poor substitute for true accessibility because it is entirely unidirectional. Allowing Deaf users to communicate natively with technology requires treating sign language as a primary language format. By attaching real commercial value to this data and compensating the communities that generate it, Talksign is laying the groundwork for a more equitable AI infrastructure.

It is exactly the kind of structural thinking the broader tech ecosystem desperately needs to adopt.

Also read: TalkSign: Edidiong Ekong just wanted to talk to friends; now he is using AI to bridge the silenceThe Marketplace is now live. If you hold sign language data and want to list it, if you’re a signer or organisation interested in contributing or annotating, or if you’re building something that needs licensed multimodal sign data, Talksign would like to hear from you. You can sign up here: https://marketplace.talksign.co/ or reach them directly at [email protected].