Age estimation based on children's voice: a fuzzy-based decision fusion strategy

Area of Science:

• Speech Analysis
• Machine Learning
• Signal Processing

Background:

• Automatic speaker age estimation is a complex challenge in speech analysis.
• Existing methods face difficulties in accurately determining age from vocal characteristics.

Purpose of the Study:

• To present a novel, accurate approach for automatic speaker age estimation.
• To leverage a "divide and conquer" strategy using vowel classes and fuzzy data fusion.

Main Methods:

• Speech data were segmented into six groups based on vowel classes.
• Mel-frequency cepstral coefficients (MFCCs) were extracted for each group.
• Single-layer feed-forward neural networks (SLFNs) and self-adaptive extreme learning machines (ELMs) were used for classification, followed by fuzzy data fusion.

Main Results:

• The proposed method demonstrated significant improvements in age estimation accuracy, reaching up to 53.33%.
• Fuzzy fusion effectively aggregated complementary age-related information from different vowel classes.
• The approach outperformed several state-of-the-art age estimation techniques.

Conclusions:

• The "divide and conquer" strategy combined with fuzzy fusion enhances speaker age estimation accuracy.
• Vowel-class-based speech segmentation and fusion of classifier outputs are effective for age estimation.
• This method offers a robust solution for automatic speaker age determination across various age groups, including children.