Analysis for speech recognition Using artificial intelligence methods (RATTLE)

Overview:

In this work, we are collecting and analyzing samples of voices characterized by varying degrees of laryngeal pathology in order to understand how to improve well-being and communication in individuals rendered frail by such pathologies. We employ innovative artificial intelligence techniques for voice analysis and characterization, evaluating the effectiveness and limitations of current technologies in use. Systems in use at the state of the art are based on advanced linguistic models (developed through machine learning) that effectively recognize non-pathological voices, with a high degree of robustness, allowing the recognition of voices partially affected by background noise and other linguistic artifacts (e.g., regional inflections). In contrast, the accuracy of state-of-the-art systems decreases rapidly in the case of voices affected by pathologies that alter the regular harmonic structure of the voice. In particular, we consider subjects who have undergone partial laryngectomies. The aim of the project is to make it possible to assess the state of the art in the recognition of pathological voices, generating guidelines for returning fragile subjects to a more inclusive access to a whole series of systems that have achieved “pervasive” diffusion in the most varied application domains in which the voice is recognized and transformed for control, implementation and communication.

Our data collection campaign page: https://speakai.projects.unibz.it/

in collaboration with Ospedale Cannizzaro https://www.ospedale-cannizzaro.it/ 

Funding body: Fondazione Pfizer  https://www.fondazionepfizer.it/

Outputs:

---

---