The paper 'Disentangling Depression from Cognitive Decline in Elderly Speech Using Concurrent Clinical Assessments', by Jeon Woo-ri (master's student) and Professor So Jung-min (corresponding author) of the Intelligent Connected Systems Laboratory (ICSL), which presents a way of isolating and detecting only depressive symptoms in the speech of older adults with mild cognitive impairment, has been accepted for publication as a regular paper at Interspeech 2026.

The speech of older adults with mild cognitive impairment (MCI) carries acoustic changes from cognitive decline together with acoustic changes from depression. If the two factors are not separated, a classification model risks learning patterns of cognitive decline rather than depression.

The team built a Korean elderly speech corpus, collecting speech from 89 older MCI speakers over three years and administering clinical assessments of depression (SGDS) and cognition (MMSE) at the same time points. Linear mixed-effects models and partial correlation analysis both gave the same result. Among seven groups of acoustic features, only formant features remained significantly associated with depression after controlling for cognitive function, while the F0-family features widely used for depression detection showed no significant signal. Notably, just 12 features consisting of F1 and F2 alone achieved a UAR of 0.760, outperforming both the full eGeMAPS feature set and self-supervised representations.

The research shows that when clinical assessments are obtained concurrently with speech, interpretable depression-specific acoustic markers can be found even in a population where cognitive decline and affective symptoms occur together.

[References]

Laboratory: https://icslsogang.github.io/

Conference link: https://interspeech2026.org/en-AU