A New Approach For Age Estimation System Based on Speech Signals

Developing technology and innovations have led to the development in many areas, and the age estimation with the human voice is a research area that has increased its popularity recently. For security problems or in the advertising sector, age recognition applications with voice have been used. Sound is structurally complex, but it has been seen that it is possible to extract the characteristic features of the sound. The designed system was created without giving any gender information in order to estimate age from human speech. The most popular audio feature extraction methods are Mel-Frequency Cepstrum Coefficient (MFCC) and Perceptual Linear Prediction (PLP) which were used in this study. In addition, Chroma features were also used. This study, it is aimed to get the highest efficiency from the voice features by using different feature extractors and rearranging the dataset according to the feature importance priority. For this purpose, eight age groups were formed from a dataset containing different speakers and so, the MLP (Multi-Layer Perceptron) classification method was used. Mozilla Open-Source Dataset was used in our system, and the highest accuracy rate of age classification was observed as 94.34% being the highest score in the literature.