Age Estimation based on MFCC Speech Features and Machine Learning Algorithms

This paper analyses how well various machine learning algorithms perform when used to infer a person's age from speech in real-world applications including human machine interfaces (HMI), automatic speech recognition (ASR), and interactive voice response (IVR) systems seen in call centers. The difficulty of the present techniques for extracting salient high-level speech features and classification models makes it a difficult challenge in speech processing to categorize speakers according to their age. We combine PCA with machine learning methods including Support Vector machine (SVM), Decision Tree (DT), and Random Forest (RF) to create a novel age estimation strategy to address these issues. These methods, each using a distinct PCA and MFCC, are examined to learn more about how performance relates to the features taken from the voice corpus. Mozilla's Common Voice dataset, a free and crowd sourced voice corpus, was used as the speech corpus for the tests. The results for age categorization utilizing PCA, cross validation, and seeding in training and testing using different machine learning approaches are pretty good. Better accuracy compared to standard classification is provided by the best features chosen by PCA. The outcomes hold promise for usage in automatic speech recognition and interactive voice response systems that interface with humans.