Automatic Breathing Pattern Analysis from Reading-Speech Signals.

Area of Science:

• Biomedical Engineering
• Speech Science
• Physiological Monitoring

Background:

• Speech production is intrinsically linked to the breathing process.
• Breathing patterns offer insights into psychological, physiological, and cognitive states.
• Current methods for capturing breathing patterns require costly and uncomfortable equipment like respiratory belts.

Purpose of the Study:

• To develop a method for extracting breathing patterns directly from speech signals.
• To investigate the feasibility of using smartphone microphones for non-invasive respiratory monitoring.
• To identify distinct breathing patterns and categorize them based on speech signals.

Main Methods:

• Simultaneous speech and breath signals were recorded from 100 Indian participants (20-25 years old).
• Participants read a phonetically balanced English passage.
• Time-domain features and a regression network were employed to extract breathing patterns from speech.
• Computational modeling was used for classification of breathing categories.

Main Results:

• Five distinct breathing templates were identified, categorized into two broad speech-breath patterns.
• Breathing patterns were extracted from speech with a Pearson correlation coefficient of 0.70 for one category.
• The study achieved a classification accuracy of 79% in distinguishing between the two breathing categories using computational modeling.

Conclusions:

• Breathing patterns can be effectively extracted from speech signals, offering a non-invasive alternative to traditional methods.
• Speech signal analysis holds potential for remote and accessible physiological state monitoring.
• The identified breathing categories and extraction methods provide a foundation for further research in speech-based health assessment.