An educational project driven by three vocational training institutes in Alicante and Elche has grown into one of the most ambitious initiatives in the country to improve communication for individuals with speech disorders using artificial intelligence. Four years since its launch, the LARA project has gathered more than 43,000 voice samples from users with speech difficulties caused by various pathologies or their aftereffects. This unprecedented database is now being used to train algorithms capable of interpreting what these individuals want to say.
The initiative originated in the province of Alicante, spearheaded by the Gran Via Institute in Alicante, alongside the Severo Ochoa and Victoria Kent Institutes in Elche. It has since expanded to a national level after securing over 100,000 euro in funding from the Ministry of Education. Currently, the Poligono Sur Institute in Seville and the Albarregas Institute in Merida are also participating in the scheme.
“Before, we only had an application to capture voices. Now we have developed a second phase, an output application, because artificial intelligence is now able to recognise patterns thanks to the more than 43,000 voice samples we have managed to collect,” explains Emma Sancho, a teacher at IES Gran Via and one of the project coordinators.
The goal is to address a reality that affects thousands of individuals. Current voice assistants and recognition systems work correctly when diction is clear, but they encounter enormous difficulty understanding users with cerebral palsy, stroke sequelae, neurological diseases, or any disorder that affects pronunciation.
To address this problem, students from various educational levels are working alongside associations, occupational centres, and speech therapy professionals. Participating users regularly record sentences designed to capture different phonemes and speech patterns.
Each recording is stored in the cloud, with the express authorisation of the participants, and then undergoes a data cleaning and processing procedure. “The more samples we have, the better we can train the artificial intelligence and reduce any biases that may appear,” Sancho points out.
The system has already proven capable of identifying what a person means, even when their speech is difficult for someone outside their immediate environment to understand. Currently, the result is displayed in written format, although the ultimate goal is to generate a real-time voice response in the future. The project’s philosophy is summarised in the motto that has guided it since its inception: “Communication is a right, not a privilege.”
The promoters of LARA aim to enable people with speech difficulties to function more independently, without constantly needing to rely on others to make themselves understood.
“Often these people need pictograms or the help of someone to interpret what they want to communicate. What we aim for is that they can relate with greater autonomy because cognitively they have all the capabilities, but they encounter barriers in communication,” Sancho points out.
Volunteer users who contribute their voices through a series of phrases help create a database of voices associated with different phonemes. Subsequently, subsets of these audio recordings are selected to train different models to work better with a specific group or organisation. The project’s main challenge is enabling the model to generalise and learn to interpret the voice of another user based on one user’s voice.
Despite the progress made, those in charge acknowledge that the future of the initiative now depends on the involvement of companies capable of transforming the work developed into a fully operational tool. Public funds have been allocated to technological equipment, travel, training activities, and technical development. However, educational institutions cannot single-handedly bear the costs of maintaining servers, scaling the technology, or marketing the application.
“The LARA project is educational. We need public or private companies to come in and continue the work we’ve done because we can’t maintain all the necessary infrastructure,” Sancho warns. Meanwhile, data collection continues. Far from being over, the project’s promoters believe this phase is never truly finished, as more data inevitably means better results in artificial intelligence.
Other Articles
Woman arrested after stabbing in Villajoyosa
