Phoneme Based Model for Gender Identification and Adult-Child Classification
Mohan Bansal, Pradip Sircar · 2019
Speech is the primary way of communicating messages from a speaker to a listener. The listener can identify details about the speaker's personality, age, gender, accent, emotions from the voiced signal. In this paper, we address the gender identification and adult-child classification using the phoneme-based parametric model. The multi-component amplitude and frequency modulated (AFM) signal model is suitable for the entire duration of a phoneme, and the estimation of model parameters is robust to various recording conditions. The gender identification and adult-child classification are done using the features extracted from the phoneme-based AFM signal model and applying the support vector machines (SVM) with the k-fold cross-validation (CV). The experimental results have been performed in the various recording conditions and show that the proposed technique is suitable for gender identification and age classification independent of spoken text, emotion of the speaker, environment and microphone set-up. It is demonstrated that the proposed model-based technique performs with a high probability of correct identification/classification.