Voice recognition and aphasia: can computers understand aphasic speech?

Purpose: This preliminary investigation of the use of voice recognition software by people with aphasia presents a new method of training the software, which allows the aphasic user to by-pass the linguistically demanding standard training. It examines how accurately voice recognition software performed with aphasic speech and language as compared to that of control subjects. Method: Five non-impaired controls and six participants with aphasia used a vocabulary of 50 words and 24 phrases to train the software over a maximum of five sessions. For the people with aphasia baseline assessments provided a profile of speech and language impairment. Measures of software accuracy were taken for all subjects at baseline and after four subsequent training sessions, for both single word and phrase level dictation. Results: Word level production resulted in similar accuracy levels for both groups. Phrase level production showed greater accuracy levels than single word level production for both aphasics and controls, but participants with greater speech difficulties had lower software accuracy scores. Conclusions: Training the software on a specific vocabulary allows people to access it whose speech and language difficulties would otherwise have prevented them. Findings are discussed in relation to use of the software as a dictation tool and as an input device to therapy software.