Disentangling Depression from Cognitive Decline in Elderly Speech Using Concurrent Clinical Assessments

Woori Jeon; Seunghee Ha; Sang-Kyu Lee; Ji Hye Yoon; Tae-Jin Yoon; Seung Jin Lee; Woojae Han; Jungmin So

eess.ASOct 15, 2026

Disentangling Depression from Cognitive Decline in Elderly Speech Using Concurrent Clinical Assessments

Authors: Woori Jeon, Seunghee Ha, Sang-Kyu Lee, Ji Hye Yoon, Tae-Jin Yoon, Seung Jin Lee, Woojae Han, Jungmin So

Organizations: Sogang University, South Korea · Hallym University, South Korea · Sungshin Women’s University, South Korea

Abstract

Detecting depression from speech in elderly patients with mild cognitive impairment (MCI) is complicated by overlapping acoustic effects of cognitive decline. Without disentangling these, classifiers risk learning cognitive rather than depression-specific patterns. We present a Korean elderly speech corpus of 89 MCI speakers collected over three years with concurrent depression (SGDS) and cognitive (MMSE) assessments. Both linear mixed-effects models and partial correlations identify the same pattern: of seven feature groups, only formant features are associated with depression after controlling for cognitive function, while widely used F0 features carry no signal. The F1+F2 subset achieves UAR 0.760 with just 12 features, outperforming both the full eGeMAPS baseline and self-supervised representations. These findings show that concurrent clinical assessments can identify interpretable, depression-specific acoustic markers in populations where cognitive and affective symptoms co-occur.

Figures & tables

Year	N	Dep.	Non-d.	Age	MMSE	SGDS
Y1	61	14	47	77.7 $\pm$ 5.8	23.8 $\pm$ 3.3	4.5 $\pm$ 3.9
Y2	72	13	59	78.5 $\pm$ 6.1	23.2 $\pm$ 3.5	4.0 $\pm$ 4.1
Y3	76	14	62	78.1 $\pm$ 6.5	23.4 $\pm$ 3.2	4.6 $\pm$ 3.3
Total	209	41	168	78.1 $\pm$ 6.2	23.5 $\pm$ 3.3	4.4 $\pm$ 3.8

Table 1: Corpus statistics. Dep. = depressed (SGDS

\geq

8).

Group	Low-level descriptors	#
F0	F0 (semitone), rising/falling slope	10
Energy	Loudness, peaks/sec, equiv. sound level	12
Spectral	Alpha ratio, Hammarberg idx, slope,	17
	spectral flux (V/UV)
MFCC	MFCC 1–4 (all frames + V)	16
Voice qual.	Jitter, shimmer, HNR, H1–H2, H1–A3	10

Table 2: Seven feature groups derived from the 88 eGeMAPS features. Each LLD is expanded into mean and standard deviation over voiced (V), unvoiced (UV), or all frames, yielding the total count per group. Temporal features are already computed per utterance, so no mean/standard deviation is applied.

			LMM		Pcorr
Group	$d$	%PC1	$\beta$	$p$	$r$	$p$
F0	10	40	$+.005$	$.894$	$+.049$	$.478$
MFCC	16	24	$+.025$	$.432$	$+.092$	$.188$
Temporal	5	54	$-.034$	$.281$	$-.127$	$.066$
Voice qual.	10	38	$+.009$	$.822$	$-.039$	$.573$
Energy	12	74	$-.097$	$.023^{*}$	$-.061$	$.384$

Table 3: Group-level analysis: LMM SGDS coefficient and partial correlation with SGDS controlling for MMSE, age, and gender, applied to PC1 of each group.

{}^{*}p<.05

Baselines
Method	$d$	UAR	Sens	Spec	Acc	F1
eGeMAPS full	88	.655	.512	.798	.742	.438
CER + WER	2	.645	.683	.607	.622	.415
SSL + SVM
WavLM-Base	768	.662	.634	.690	.679	.437
wav2vec2-Base	768	.674	.585	.762	.727	.457

Table 4: Depression classification (LOSO CV,

N

=209). † Linear head (5-seed average); all others SVM RBF

C

=0.1. Best UAR in bold.

Subgroup	$d$	UAR	Sens	Spec	Acc	F1
F1	6	.618	.634	.601	.608	.388
F2	6	.747	.756	.738	.742	.534
F3	6	.641	.585	.696	.675	.414
F1+F2	12	.760	.805	.714	.732	.541
F1+F3	12	.662	.634	.690	.679	.437
F2+F3	12	.735	.707	.762	.751	.527

Table 5: Formant subgroup classification. Same setup as Table 4 . Best UAR in bold.

Depression Detection

Depression detection uses language, behavior, speech, and biological signals to identify depressive symptoms and estimate their severity.

See all 77 papers

Learning to Attend to Depression-Related Patterns: An Adaptive Cross-Modal Gating Network for Depression Detection
Apr 11, 2026Hangbin Yu, Yudong Yang, Rongfeng Su +2Depression Detection Clinical Speech Processing
Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, China · University of Chinese Academy of Sciences · Key Laboratory of Biomedical Imaging Science and System, Chinese Academy of Sciences, China
Reading between the Lines: Leveraging Large Language Models for Global Dementia and Depression Assessment from Clinical Interviews
Jun 16, 2026Franziska Braun, Alea Rüggeberg, Thomas Ranzenberger +4Depression Detection Mental Health
1TH Nürnberg · 2FAU Erlangen · 3PMU Klinikum Nürnberg, Germany
Depression Markers in Speech: An Approach based on Tract Variables Dynamics
Jul 28, 2026Sahar Altalhi, Tanaya Guha, Alessandro VinciarelliDepression Detection Speech Processing
University of Glasgow (United Kingdom) and Taif University (Saudi Arabia) · University of Glasgow (United Kingdom)

Apr 11, 2026J/K move · Enter open · S save

Learning to Attend to Depression-Related Patterns: An Adaptive Cross-Modal Gating Network for Depression Detection

Hangbin Yu, Yudong Yang, Rongfeng Su, Nan Yan+1

Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, China · University of Chinese Academy of Sciences · Key Laboratory of Biomedical Imaging Science and System, Chinese Academy of Sciences, China

Depression Detection Clinical Speech Processing

Open paper AskPDF arXiv

Automatic depression detection using speech signals with acoustic and textual modalities is a promising approach for early diagnosis. Depression-related patterns exhibit sparsity in speech: diagnostically relevant features occur in specific segments rather than being uniformly distributed. However, most existing methods treat all frames equally, assuming depression-related information is uniformly distributed and thus overlooking this sparsity. To address this issue, we proposes a depression detection network based on Adaptive Cross-Modal Gating (ACMG) that adaptively reassigns frame-level weights across both modalities, enabling selective attention to depression-related segments. Experimental results show that the depression detection system with ACMG outperforms baselines without it. Visualization analyses further confirm that ACMG automatically attends to clinically meaningful patterns, including low-energy acoustic segments and textual segments containing negative sentiments.

Abstract

Depression Detection

Explore similar work

Learning to Attend to Depression-Related Patterns: An Adaptive Cross-Modal Gating Network for Depression Detection

Reading between the Lines: Leveraging Large Language Models for Global Dementia and Depression Assessment from Clinical Interviews

Depression Markers in Speech: An Approach based on Tract Variables Dynamics