Data Collection for Automatic Depression Identification in Spanish Speakers Using Deep Learning Algorithms: Protocol...

Area of Science:

• Computational linguistics
• Mental health informatics
• Machine learning for healthcare

Background:

• Depression diagnosis relies on subjective questionnaires, lacking objective biomarkers.
• Deep learning models for voice depression classification require extensive, multilingual datasets, which are currently scarce.
• Existing voice analysis for depression is limited by data availability and language diversity.

Purpose of the Study:

• To establish a high-quality voice dataset for depression classification in Spanish speakers.
• To facilitate deep learning research by providing a novel dataset.
• To explore the utility of smartphone recordings for depression detection.

Main Methods:

• Collected voice recordings from at least 60 Spanish-speaking participants diagnosed with depression and controls.
• Utilized both professional-grade and smartphone microphones for data capture.
• Gathered depression labels via the Patient Health Questionnaire-9 and other relevant speech-influencing data.

Main Results:

• The created dataset enables immediate research into Spanish-language voice depression classification.
• Facilitates evaluation of audio quality's impact on deep learning models.
• Supports research on the practical application of voice depression classification via smartphone apps.

Conclusions:

• This work contributes to objective, automated depression classification using voice analysis.
• Addresses data scarcity and language barriers in deep learning for mental health.
• Paves the way for advanced, accessible AI-driven mental health diagnostic tools.