In this paper, the hypothesis is a hybrid intelligent model of speech-based gender and age estimation technologies based on the integration of deep learning and classical machine learning models. Gender classification is done by deep neural network trained on Mel-spectrogram representation, which is effective in extracting time-frequency features of human speech. In age estimation, a broad category of statistical and spectral examples, including spectral centroid, spectral bandwidth, zero-crossing rate, spectral roll-off, pitch and Mel-Frequency Cepstral Coefficients (MFCCs), will be derived and learned by a Random Forest classifier. The suggested system can analyze the audio files offline and receive the microphone in the real-time, therefore, being implemented in reality. Audio signal processing such as noise removal as well as normalization, silence, elimination are used to enhance the quality of the signal. It has been experimentally demonstrated that the deep learning method can significantly contribute to improving gender prediction accuracy, and the Random Forest classifier is able to differentiate various age groups. The suggested structure is very robust, scalable, and feasible and, therefore, can be implemented in speech-based biometric systems, human-computer interaction, and intelligent voice-assisted systems.
Priya et al. (Thu,) studied this question.