Intra- and Inter-database Study for Arabic, English, and German Databases: Do Conventional Speech Features Detect...

Area of Science:

• Speech processing
• Biomedical engineering
• Otolaryngology

Background:

• Voice disorders affect a significant global population, necessitating reliable diagnostic methods.
• Subjective voice evaluations are prone to human error and clinician variability.
• Objective, automated systems offer noninvasive, consistent screening for voice pathology, especially in remote areas.

Purpose of the Study:

• To evaluate the reliability of conventional speech features for voice pathology detection.
• To determine if conventional speech features correlate with voice quality.
• To assess the performance of a Mel-frequency cepstral coefficients (MFCC)-based automatic detection system across different voice disorder databases.

Main Methods:

• Development of an automatic voice disorder detection system utilizing Mel-frequency cepstral coefficients (MFCCs).
• Testing the system on three distinct voice disorder databases.
• Analysis of detection rates for both intra-database and inter-database comparisons.

Main Results:

• The accuracy of the MFCC-based system varied significantly across different databases.
• Intra-database detection rates ranged from 72% to 95%.
• Inter-database detection rates were lower, ranging from 47% to 82%.

Conclusions:

• Conventional speech features, including MFCCs, are not reliably correlated with voice quality.
• These features demonstrate limited reliability for the accurate detection of voice pathology.
• Further research into more robust features is needed for effective automatic voice disorder screening.