Human Voice Analysis to Determine Age and Gender

Age estimation from a speaker's voice is a research area that has garnered significant interest in recent years. The most popular method to extract voice features presently involves using MFCC coefficients which are the statistical features acquired from a kind of Spectral depiction of the audio clip (a nonlinear “spectrum-of-a-spectrum”). In the Mel scale, the divisions of the frequency band closely mimic the human auditory systems response. Post extraction, the needed features are extracted and the relevant classification or regression models/operations are applied. Numerous deep learning algorithms have been used so far in the studies for age and gender estimation. In this paper, the attempt is the same. A new labeled dataset is built by collecting the audio samples of people belonging to a variety of age groups and two genders (male and female) from the web. To this dataset the MFCC feature extraction program was applied; after extraction, different regression models were used to compute the age of the speakers and classification models to identify their genders.