Improving the accuracy of automated cleft speech evaluation.

Area of Science:

• Speech Pathology
• Medical Technology
• Pediatric Care

Background:

• Cleft palate speech disorders significantly impact quality of life.
• Objective evaluation of cleft speech is crucial for effective treatment and research collaboration.
• Existing methods for cleft speech analysis can be subjective and resource-intensive.

Purpose of the Study:

• To describe an updated, efficient automated cleft speech evaluator.
• To assess the accuracy of the updated evaluator in identifying resonance and articulatory errors.
• To compare the performance of the updated evaluator against previous models and human expert ratings.

Main Methods:

• Speech samples from 73 patients (60 for training, 13 for testing) were collected.
• The automated evaluator was trained using specific speech sounds and sentences.
• Evaluator performance was compared against independent ratings from two experienced speech pathologists.

Main Results:

• The updated automated cleft speech evaluator achieved 85% agreement with combined speech pathologist ratings.
• This represents a significant improvement over the previous model's 65% agreement rate.
• The system demonstrated good accuracy despite a relatively small training dataset.

Conclusions:

• The updated automated cleft speech evaluator shows high accuracy in classifying cleft speech errors.
• This technology has the potential to standardize cleft speech assessment globally.
• Further increases in training data are expected to enhance the evaluator's accuracy, potentially matching human listener performance.