Improving Speaker Recognition by Biometric Voice Deconstruction

Area of Science:

• Biometrics
• Speech Processing
• Forensic Science

Background:

• Traditional identification methods like fingerprint and facial recognition are insufficient in critical environments.
• The rise of terrorism utilizing social media necessitates alternative biometric solutions.
• Voice analysis is emerging as a crucial biometric characteristic when other methods are unavailable.

Purpose of the Study:

• To investigate the effectiveness of gender-dependent voice characterization for speaker identification.
• To explore the use of voice features derived from glottal source and vocal tract deconstruction.
• To enhance speaker recognition rates compared to classical biometric approaches.

Main Methods:

• Developing a gender-dependent model for speaker characterization.
• Extracting biometric parameters from deconstructed voice signals (glottal source and vocal tract estimates).
• Conducting experimental validation on controlled and non-controlled acoustic condition databases.

Main Results:

• Gender-dependent voice feature extraction significantly improves speaker recognition rates.
• The proposed method demonstrates effectiveness in both controlled and real-world (mobile phone network) conditions.
• Deconstructing voice into its fundamental components enhances biometric parameter distinctiveness.

Conclusions:

• Gender-specific voice analysis combined with advanced feature extraction offers superior speaker identification.
• This approach provides a robust solution for person identification in challenging environments.
• The findings support the adoption of sophisticated voice biometrics for security applications.