Study of the Performance of Automatic Speech Recognisers with Speech Impaired Children

摘要

This study examines the feasability of using automatic speech recognition devices with speech impaired children. With advice from a speech therapist, a group of five children with various speech and language problems were selected. Structured conversations between the speech therapist and individual children were recorded in an informal play setting at the children's school. Two children with reasonably normal speech were also recorded. The recordings were pre-processed, edited and annotated. Subsequently isolated word recognition experiments were performed using computer simulation. The results of the experiments suggest that current speech recognition techniques are not sufficiently powerful to accomodate the degree of variability which occurs in the speech of children with the types of speech impairment considered in the study.