Misophonia Sound Recognition Using Vision Transformer

Area of Science:

• Neuroscience
• Audio Signal Processing
• Artificial Intelligence

Background:

• Misophonia involves abnormal emotional reactions to specific sounds like eating or breathing.
• Effective sound classification is crucial for developing misophonia interventions.
• Deep learning, particularly Convolutional Neural Networks (CNNs), has shown promise in audio analysis.

Purpose of the Study:

• To propose a transformer-based deep learning algorithm for automatic misophonia trigger sound identification.
• To characterize misophonia trigger sounds using acoustic features.
• To evaluate the algorithm's accuracy and specificity in classifying these sounds.

Main Methods:

• A transformer-based deep learning model was developed for audio classification.
• The algorithm was trained to identify and characterize misophonia trigger sounds.
• Performance was evaluated based on accuracy and specificity metrics.

Main Results:

• The proposed transformer-based algorithm achieved high accuracy in classifying misophonia trigger sounds.
• The model demonstrated high specificity in identifying trigger sounds.
• Acoustic feature characterization was successfully performed.

Conclusions:

• Transformer deep learning models offer a powerful approach for misophonia sound classification.
• The developed algorithm provides a strong foundation for creating new misophonia therapies.
• Further research can leverage these findings for clinical applications.