Keia, a human solution to sign language interpretation

Sign language is way more complex than hand movement and Keia does considerate this.

General presentation

Founded in 2019, Keia is an accessibility solution dedicated to people with disabilities. Today, nearly 7% of the world’s population — about 460 million people according to the WHO — requires a hearing aid device. When a human interpreter is not available, Keia offers an artificial interpretation technology that automatically translates any audio, video or written content into signed language (Sign Language, French Complete Spoken Language and subtitles) through a 3D avatar.

To understand the interest of their approach and to break the preconceived ideas about deafness and sign language, we invite you to take a look at their FAQ

Why not just use subtitles?

Today, due to a bilingual education that is not sufficiently provided, there is a strong functional illiteracy among the deaf. For the overwhelming majority of them (we speak about 70%) reading is not easy. Thus, a majority of deaf signers in France and elsewhere do not quite understand what you write.

The technology

Study project or personal conception, we see more and more posts on the networks putting forward a code allowing to “interpret” the sign language by using artificial intelligence, but what does Keia really bring? To understand it, we have to look at what the sign language is, and what it is not.

As the startup says in its tweet above, sign language is not one sign per letter, it is not soft, it is not emotionless. Just like spoken languages, sign language does not stop at words (or rather signs in this case), you have to take into account facial expressions, gestures, hand placement, body placement. Keia proposes just that, an avatar interpreting content in sign language, with a human aspect, and not only by a sequence of characters or words.

To do this, the startup uses a deep learning algorithm that performs an initial translation of the content sent. The translation is then analyzed by an expert (human) who certifies and improves, if necessary, the work of the AI. Once the translation is certified, you can select the 3D avatar of your choice to interpret it.

Requiring the intervention of a human expert, the translation is not as instantaneous as one might wish. However, this seems like a small price to pay for a good quality final output that will be much more enjoyable and inclusive for the people involved.

Offer

Bilingual LSF/French data or corpora do not exist. Keia creates them from scratch, hand in hand with translators and experts in LSF on the basis of various customer contents. Thus, Keia does not offer an instantaneous translation solution, but accompanies companies in the definition of their business vocabulary, the translation of their content, the design of their avatar, the integration of their translations as well as the communication of their approach.

Support / Clients / Partners

The French government recently equipped itself with ANAE, a digital assistant created by Keia to animate the “Vaccine Pass “ and “Vaccines “ pages of the Covid-19 space.

Keia’s other clients include Macif, Microsoft, Inclood and Cora.

Keia has been/is also supported by many structures such as the agency Paris&Co via the HUA program, Furahaa, the foundation for Hearing, the SNCF, the national federation of deaf people in France, Viva Technology, Open Mind,…

Target

In 2019, the WHO estimates that there are 460 million deaf people in the world, including 34 million children.

By 2050, the WHO estimates that 900 million people will suffer from disabling hearing loss.

Moreover, in France, there are approximately 400 qualified interpreters, a number that is still insufficient to meet the needs of the deaf population. The number of qualified sign language translators is even more limited.

Competitors

Keia welcomed the initiative of innovators and students, who seek to bring an innovative solution to improve the accessibility of Deaf people in their language, in Sign Language.

However, these technologies do not allow to translate the sign language in a realistic way, it is only the signs that have almost no movement or the manual alphabet. The signs are linked according to a complex grammar that goes beyond hand recognition.

On our side, we have not found a direct competitor to Keia. Researchers from the University of Surrey in England have developed SignGAN: a program that generates, in real time, a video of a person signing everything he says.