Automatic Speaker Recognition System Based on Gaussian Mixture Models, Cepstral Analysis, and Genetic Selection of...

Area of Science:

• Computer Science
• Signal Processing
• Biometrics

Background:

• Speaker recognition systems face challenges in open-set identification and verification under noisy conditions, such as telephone transmission.
• Existing systems often struggle with performance degradation due to varying recording quality and background noise.

Purpose of the Study:

• To present a novel Automatic Speaker Recognition System (ASR System) designed for robust speaker identification and verification.
• To detail the architecture and internal processing modules of the proposed ASR System.
• To evaluate the system's performance against competing systems on certified voice datasets.

Main Methods:

• The ASR System employs genetic algorithms for feature selection and internal parameter optimization.
• Proprietary feature generation techniques are utilized.
• Gaussian mixture models are used for the classification process.

Main Results:

• The ASR System demonstrated improved speaker identification and verification results compared to existing systems.
• The system achieved high performance on a certified voice dataset, particularly under simulated telephone transmission conditions.
• The dual application of genetic algorithms proved effective in enhancing system accuracy.

Conclusions:

• The developed ASR System offers a significant advancement in speaker recognition technology, especially for telephone-based applications.
• The integration of genetic algorithms and Gaussian mixture models provides a robust and effective approach to speaker recognition.
• The system's performance highlights its potential for real-world applications with challenging voice data.