Auditory perception based system for age classification and estimation using dynamic frequency sound

摘要

Human age is a crucial factor in social interaction. It determines the way we interact with others. It is also a relevant forensic issue that can provide helpful information in legal and criminal investigations. Thus, human age estimation has a wide range of real-world applications related to human computer interaction and forensic sciences. Based on auditory perception, in this paper, we investigate a new biometric trait for human age classification and estimation. For this purpose, several techniques of Machine Learning, including Random Forests (RF), Support Vector Machines (SVM), Linear Regression (LR), Polynomial Regression (PR), Ridge Regression (RR) and Artificial Neural Networks (ANNs), are used to estimate the age of the volunteers. To evaluate the performances of our experiment, a dataset of 837 tests have been collected with different ages ranging from 6 to 60 years. The results show a good accuracy between 86% and 92% of reasonable classification and 98.2% of good age estimation with a root-mean-square error of 2.6 years. Results are found to be significant and show that auditory perception is one of the practical interests in real-world applications. The dataset we used will be made publicly available online.