Talksign has launched the Talksign Marketplace, a platform where Deaf communities, researchers, institutions, and companies can contribute, annotate, license, and access high-quality sign language datasets. The company positions the new platform as the foundational infrastructure layer required to build artificial intelligence that genuinely works for Deaf people.

The launch tackles a long-standing gap in AI development. 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; it is a severe lack of structured, licensable data.

While leading speech models train on decades’ worth of scraped open-web audio, 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 resulting products inevitably fail in the real world, struggling with regional dialects, native signing rhythms, and the complex facial grammar required to convey intensity or ask a question.

Edidiong Ekong, Talksign’s founder and CEO, said in an exclusive statement shared with Technext: “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.”

“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 added.

The Scale of the Gap

According to World Health Organisation data, over 430 million people globally experience disabling hearing loss — a figure the WHO projects will exceed 700 million by 2050 — and 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. They are also mutually unintelligible — American Sign Language and British Sign Language differ fundamentally despite both countries sharing spoken English — so there is no single universal “sign language” for a model to master. Building AI that understands them requires solving a genuinely difficult multimodal computing problem, with models that must reason across multiple visual channels in real time.

The data needed to train such 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.

How the Marketplace Works

Talksign’s marketplace functions across four core pillars designed to commercialise and standardise sign language 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 significant 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.

A Blueprint for African AI

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 that locks them out of the global AI economy. The World Federation of the Deaf estimates that more than 80 percent of the world’s roughly 70 million deaf people live in developing countries, a distribution that gives regions like Africa outsized weight in any effort to build representative signing data.

African technologists building natural language processing tools for languages such as Hausa, Yoruba, or Swahili consistently hit a data wall. The Talksign framework offers a 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.

Consent, Verification, and Biometric Data

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

Sign language data presents an additional challenge. Because it is inherently biometric, capturing a person’s face and body, it cannot be anonymised like text. It also sits in the most heavily regulated tier of personal data: the EU’s GDPR, for example, classifies biometric data processed to identify a person as a “special category” requiring explicit consent, and the EU AI Act layers further rules on biometric systems. 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.

Proof of Concept

Talksign claims the infrastructure was born out of its own internal development hurdles, and 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.

Beyond Captioning

Captioning, Talksign argues, 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 sign language data and compensating the communities that generate it, the company says it is laying the groundwork for a more equitable AI infrastructure — exactly the kind of structural thinking the broader tech ecosystem needs to adopt.

The Marketplace is now live. Holders of sign language data who want to list it, signers and organisations interested in contributing or annotating, and builders who need licensed multimodal sign data can sign up at marketplace.talksign.co or reach the team directly at sales@talksign.co.