RAPNIC Project: Automatic Recognition of Unintelligible Speech in Catalan
RAPNIC Project: Automatic Recognition of Unintelligible Speech in Catalan
RAPNIC is a project focused on developing and training an Artificial Intelligence (AI) model capable of recognizing unintelligible speech in Catalan.
(2025-2026)
Despite technological advancements, voice recognition systems still face significant challenges in identifying and interpreting speech patterns that deviate from standard norms, particularly in the case of so-called “unintelligible speech.” As a result, people with speech disorders, such as dysarthria, often struggle to be understood and are unable to access technological tools that could facilitate their communication and autonomy. Furthermore, the few existing solutions are primarily available in English, creating an additional barrier for speakers of other languages, including Catalan.
The RAPNIC project (Automatic Recognition of Unintelligible Speech in Catalan) aims to address this need by developing an AI-powered solution capable of recognizing and interpreting unintelligible speech in Catalan. This technology will allow individuals with speech disorders to interact effectively with voice assistants, speech-to-text transcription systems, and other digital tools, thereby improving their autonomy and quality of life.
What is RAPNIC?
RAPNIC is a pioneering initiative by the iSocial Foundation that seeks to eliminate a major technological barrier for people with speech disorders. Using Artificial Intelligence (AI), RAPNIC aims to develop a system capable of recognizing and understanding unintelligible speech, which current voice recognition systems fail to comprehend.
To achieve this, the project plans to create a database of unintelligible Catalan speech, built from voice recordings of individuals with speech disorders, with a particular focus on dysarthria— a motor impairment affecting facial muscles and making the pronunciation of certain phonemes difficult. This database will train AI algorithms using deep learning technology to recognize the sound patterns of these speech variations.
In the project’s initial phase, work will be carried out on a corpus of recordings from individuals with Down syndrome and cerebral palsy, as their speech patterns generally present fewer alterations than other forms of dysarthria, making AI training more manageable. The database will contain at least 100 hours of recorded speech and involve the participation of 120 volunteer speakers with Down syndrome and cerebral palsy, along with social professionals, speech therapists, deep learning specialists, and computational linguists.
The ultimate goal of RAPNIC is to extend this system, first to the 22,000 people with Down syndrome and cerebral palsy within the Catalan linguistic community and, in a later phase, to the entire population of 49,000 individuals with speech disorders in the Catalan-speaking region, including those with more severe conditions. RAPNIC will enable affected individuals to communicate more effectively with those around them, give voice commands to digital assistants, transcribe their speech into text, and interact with other digital services more independently.
CSC Impulsa 2024 Award
The RAPNIC Project has been recognized as the winner of the CSC Impulsa 2024 Awards in the category of “Innovative AI Projects in the Social Sector.”
This recognition, granted by the Catalan Health and Social Consortium (CSC), includes financial support of €20,000 and technical guidance from the consortium, which will be crucial in transforming this initiative into a tangible reality.
Project Goals
RAPNIC aims to achieve the following key objectives:
- Develop an open-access dataset of unintelligible Catalan speech, based on voice recordings from individuals with speech disorders, particularly those with Down syndrome and cerebral palsy.
- Create an Artificial Intelligence model for the recognition of unintelligible speech, improving accessibility and digital inclusion for people with speech disorders.
- Launch an online application that allows individuals with speech disorders to test the model and assess its accuracy in transcribing their speech.
- Make the open dataset available to technology companies, enabling existing Catalan speech recognition tools to integrate the recognition of unintelligible speech.
- Publish a scientific study detailing the data collection process and research outcomes. This study will be presented at an international Natural Language Processing (NLP) conference, contributing to the dissemination of knowledge and methodologies used in the project.
Expected Impact and Benefits
RAPNIC has significant potential to improve the lives of people with speech disorders by enhancing their interaction with digital technologies and fostering social inclusion. The main expected benefits include:
- Increased autonomy for individuals with speech disorders, especially the 22,000 Catalan speakers with mild speech impairments, who will gain access to voice assistants, speech captioning tools, and automatic transcription services currently unavailable to them.
- Improved accessibility and communication for affected individuals, allowing them to use the same digital tools as the rest of the population, with assurance that their specific needs will be addressed.
- Reduction of the digital divide affecting people with speech disorders, particularly those with additional motor impairments.
- Creation of a speech dataset that will enable researchers to continue training open-source AI models in Catalan and help companies develop speech recognition tools that support unintelligible speech.
- Development of an AI model and methodology that could, in the future, be applied to more severe speech disorders (e.g., stroke, brain injuries) and adapted to other languages.
Leads:
Partners:
With the support of:
Projectes d’innovació
NIDUS Mental Health
Solidigital
DigitaliSSB (Promotion of the Digitalization of the Basic Social Services of Catalonia)
Vincles Alt Pirineu-Aran
Rehab-Lab
