Speaker profiling by extracting paralinguistic parameters using mel frequency cepstral coefficients

Speaker profiling is invincibly required to solve cases such as kidnapping, robbery, black mail calls, hoax, bomb threat calls and false alarms too where the evidence is in the form of telephonic conversations, tape recording, and digital recordings of speeches. Ranking them according to objective criteria such as gender, age, height and weight will be really useful. In this area many different methods of feature extraction like cepstral coefficients and world-class frequencies have been used. In this paper the first 13 coefficients of the Mel Frequency Cepstral Coefficients are used as features of a speech sample. Regression modeling was performed on these features for age, height and weight prediction. This yields mean errors which are acceptable for an application. Gender prediction was modeled with an SVM classifier and produced satisfactory results.