Evaluation of an Automatic Speech Recognition Platform for Dysarthric Speech

Area of Science:

• Assistive technology
• Speech-language pathology
• Human-computer interaction

Background:

• Commercial automatic speech recognition (ASR) software faces challenges with dysarthric speech, particularly when co-occurring with physical disabilities.
• A novel mobile and personal speech assistant (mPASS) platform was developed, utilizing speaker-dependent ASR to address these limitations.

Purpose of the Study:

• To evaluate the performance and recognition accuracy of the mPASS platform.
• To compare mPASS accuracy against commercial speaker-independent ASR.
• To explore the correlation between dysarthria severity and recognition accuracy, and user perceptions of mPASS.

Main Methods:

• Fifteen individuals with dysarthria and 20 without recorded words and sentences.
• Recognition accuracy of mPASS was compared to commercial ASR.
• Usability of mPASS was assessed using the Technology Acceptance Model (TAM) questionnaire.

Main Results:

• mPASS demonstrated significantly higher recognition accuracy for both words and sentences in individuals with and without dysarthria compared to commercial ASR.
• The mPASS platform achieved high ratings for usefulness and ease of use based on the TAM questionnaire.

Conclusions:

• The mPASS platform is a practical and accurate solution for individuals with dysarthria.
• The technology shows realistic applicability for improving communication for users with speech impairments.