An Efficient Deep Learning Based Method for Speech Assessment of Mandarin-Speaking Aphasic Patients

Area of Science:

• Computational linguistics and artificial intelligence in speech pathology.
• Machine learning applications for clinical assessment.

Background:

• Speech assessment is crucial for aphasia rehabilitation.
• Mandarin speech clarity relies on articulation, fluency, and tone.
• Efficient automatic assessment of these features is needed for aphasic patients.

Purpose of the Study:

• To present a standardized, machine learning-based method for automatic speech lucidity assessment in Mandarin-speaking aphasic patients.
• To map the relationship between speech lucidity features and aphasia severity using AI.

Main Methods:

• Utilized a convolutional neural network (CNN) with high-resolution time-frequency images.
• Adopted the Chinese Rehabilitation Research Center Aphasia Examination (CRRCAE) standard.
• Mapped speech clarity features (articulation, fluency, tone) to aphasia severity.

Main Results:

• Demonstrated statistically significant linear correlations between CNN model outputs and patient scores for articulation (0.71), fluency (0.60), and tone (0.58).
• The method effectively predicts the severity of impaired Mandarin speech.

Conclusions:

• The proposed method shows efficacy in assessing Mandarin aphasic speech severity.
• This AI-driven approach can assist speech-language pathologists and potentially improve healthcare efficiency.
• The framework is adaptable for assessing speech impairments in other languages.