# Scalable depression monitoring with smartphone speech using a multimodal benchmark and topic analysis

## TL;DR

This study explores using smartphone speech data and AI models to monitor depression severity, finding that language models paired with topic analysis can effectively track symptoms.

## Contribution

The paper introduces a scalable method for depression monitoring using speech and topic modeling, validated with a large dataset.

## Key findings

- Sentence-embedding models outperformed acoustic and lexical baselines in predicting depression scores.
- BERTopic identified six themes in speech, with 'Distress & care' showing highest depression scores.
- LLM embeddings with topic analysis offer a scalable route for ecologically valid depression monitoring.

## Abstract

Objective, scalable biomarkers are needed for continuous monitoring of major depressive disorder. Smartphone-collected speech is promising, yet clinically useful signals remain elusive. We analyzed 3151 weekly voice diaries from 284 German-speaking adults (128 MDD, 156 controls) to predict Beck Depression Inventory (BDI) scores. Sentence-embedding models outperformed lexical and acoustic baselines: Qwen3-8B achieved MAE 4.65 and R2 0.34, and stacked generalization of multilingual-E5 with Qwen3-8B further improved performance (MAE 4.37, R2 0.41). Audio embeddings added little incremental value. In an MDD-only analysis, multilingual-E5 was the top single modality (MAE 6.74, R2 0.20). To aid interpretation, BERTopic uncovered six coherent themes; BDI scores were highest for “Distress & care”, supporting clinical face validity. Together, LLM embeddings paired with lightweight topic analysis capture the dominant signal of depression severity in everyday speech and offer a scalable route to ecologically valid digital phenotyping.