Statistical differences in speech acoustics of major depressed and non-depressed adolescents

Speech acoustic parameters have been reported to provide highly efficient diagnostic cues for depression. Current studies have been focused on characterizing the speech of adult patients but it is known that adult speech differs significantly from adolescent speech, and that the onset of depression is likely to occur during adolescence itself. This paper investigates and explains the differences in speech acoustic parameters characterizing the speech of depressed and non-depressed adolescents, namely F0 (fundamental frequency), jitter, shimmer, log energy, spectral centroid, spectral entropy and glottal pulse duration. Speech data was collected from adolescents aged 14 - 18 and included 68 (49 females and 19 males) diagnosed with major depression and 71 (44 females and 27 males) diagnosed as non-depressed. The speech recordings were made during three different types of family interactions: event-planning, problem-solving and family consensus. Statistical analysis revealed that the differences between depressed and non-depressed speech acoustics strongly depend on gender and topic of conversation. For male adolescents, the glottal pulse duration was found to be the most efficient parameter in discriminating between depressed and non-depressed speech for all three interactions. For female adolescents, the fundamental frequency F0 was the most determinant during the problem solving and family consensus interactions, whilst the glottal pulse duration was the most determinant during the event planning interaction.