Screening for high risk suicidal states using mel-cepstral coefficients and energy in frequency bands

Abstract

Distinguishing high risk suicidal patients from less severely depressed patients at low risk is a critical problem. This paper describes a novel way to address this issue. The vocal characteristics of male and female speech samples from high risk suicidal and depressed patients were analyzed and distinguished using mel-cepstral coefficients and using energy in frequency bands. Two kinds of speech samples, one from an interview session and the other from a reading session, were analyzed. The results show that mel-cepstral coefficients and energy in frequency bands may be used to separate these populations and the controlled reading tended to provide better results than the interview.