Automatic assessment of pathological speech
Pathological speech differs from generally accepted speech in a specific community, being present in a wide variety situations and presenting in different variations. So far, a large number of research works have ignored the impact of non-standard speech on the speech technologies systems such as speech or speaker recognition. Automatically detecting these situations is a very important step to mitigate these effects on systems trained on standard speech.
In this context, our research group is working on the THALENTO project, aiming to progress in the use of speech and language technologies to support the evaluation of speech disorders in spanish language. This project has been designed with a multidisciplinar focus, bringing together experts from medicine & diagnosis, teaching and speech therapy and engineering. in the first stage of this project, the goal is to generate a speech corpus with different speech disorders. We are currently working on gathering all this materials to create the corpus.
Aumentative and Alternative Communication includes different systems of symbols, both gestural (mimicry, gestures or hand signals) and graphic (pictures, drawings, pictograms, words or letters). Concerning the former, pictographic systems are one of the most used alternatives when applied to people who are illiterate because of age or disability. They offer the advantage of allowing, from a basic level of communication to a very rich and advanced level, adaptable for different use cases. Our research group has collaborated in several ocassions with the Aragonese Portal of Aumentative and Alternative Communication, developers of the ARASAAC pictographic system, one of the most widely used in Spain. In this context we also highlight our participation in the IRIS european project, with the goal of developing natural interaction and communication systems.
We have an active research line focusing on the development of systems that can automatically translate texts to its pictogram equivalent. This application combines natural language processing and clustering techniques to find the pictogram that provide the closest match for a given input text.