On the Use of Recurrent Neural Network based on LSTM Model for Voice-based Age Estimation

Voice-based age estimation is an emerging technology with applications in biometric security, interactive voice response systems, and personalized recommendation services. In this paper, we explore the recurrent neural networks, in particular Long Short-Term Memory (LSTM) model for age estimation using voice. The proposed method extracts spectral features and Mel Frequency Cepstral Coefficients (MFCCs) from short voice samples, taking into consideration the gender and accent of a speaker to better estimate their age. The method was trained and tested on the Mozilla Common Voice dataset, specifically targeting English-speaking demographics across various regions. The experimental results carried out have yielded an attractive performance compared to some existing methods based on machine and deep learning algorithms, showing promising results for voice-based age estimation.