Acoustic analysis of speech based on power spectral density features in detecting suicidal risk among female patients

Abstract

Suicide is a major public health problem in the US. The procedure to measure the degree of suicidal risk in depressed patients is complicated and time consuming. Therefore, there is a need to develop a diagnostic tool to aid physicians in determining suicidal risk. Speech has been identified to be able to reflect emotional conditions including depression and suicidal thoughts. This paper represents one of the development steps in creating a speech-based diagnostic tool to help physicians to make clinical judgments. It analyzes the acoustic features based on the power spectral density extracted from the speech of female patients in order to detect suicidal risk. The focus of the experiments is on the classification between depressed and high-risk female patients. Two types of speech, spontaneous and automatic, were analyzed independently using multiple statistical approaches and the classification results are discussed. Possible outliers are observed in the spontaneous speech analysis while the automatic speech analysis produces satisfactory classification results. It is shown that speech features can be used as an indicator for suicidal risk of female patients. Potential future work that would advance our knowledge are also proposed.