SPeech ASSessment (SPASS)

Automatic speech assessment (ASA) is a technology related to automatic speech recognition (ASR) in which the goal is to automatically evaluate the linguistic correctness of a spoken word, sentence or paragraph. It is relevant in several tasks related to clinical and educational research and practice.Sinds 2010, ASR has made a giant leap forward thanks to deep learning models. The goal of SPASS is to address limitations of end-to-end (E2E) ASR models when applied to ASA. We distinguish structured and semi-structured response analysis. For structured responses, we revisit the reading tutor design of our past SBO SPACE research project. For semi-structured responses, responses by aphasia patients in conversational tasks are analysed. We characterise the responses in terms of disfluencies, lexical and grammatical mistakes.

Keywords:Speech assessment, Speech diagnostics, Fluency, Artificial intelligence, Natural language processing, Aphasia

Disciplines:Audio and speech computing, Audio and speech processing