Human Language Technologies for Speech Therapy in Spanish Language

This paper introduces Vocaliza, an application for computeraided speech therapy in Spanish language based on the use of Human Language Technologies (HLT). The objective of this application is to help the daily work of the speech therapists that train the linguistic skills of Spanish speakers with different speech impairments, working at three levels of language: phonological, semantic and syntactic. Furthermore, Vocaliza is designed to enable those who suffer speech disorders to train their communication capabilities in an easy and entertaining way, with little or no supervision once a speech therapist has configured the application for the impairment of the user. The HLT systems used in the application are Automatic Speech Recognition (ASR), speech synthesis, speaker adaptation and utterance verification. The ability of these technologies, namely ASR and speaker adaptation, to actually help users to improve their language is shown by means of the accuracy of the ASR system to detect correct and incorrect utterances according to a manual labeling of a recently acquired database containing impaired speech. The results show that accuracy reaches 87.66% when using speaker adaptation, due to its ability to model the inter speaker variability of every speaker but not their pronunciation errors.