Layer-wise Cross-Lingual Depression Detection from Speech: Analysis with Contrastive Alignment

Hugging Face Daily Papers Papers

Summary

This paper introduces CLeaD, a supervised contrastive alignment framework for cross-lingual depression detection from speech using WavLM embeddings. It reveals that previous results were inflated due to speaker identity leakage and achieves modest improvements on Mandarin speakers.

Significant disparities exist in the diagnosis and clinical presentation of depression across different linguistic populations. Speech-based depression detection performs well monolingually, but cross-lingual generalization remains an open challenge. A key reason is that prior work uses segment-level random splits without speaker grouping, leading to identity leakage that inflates reported metrics. We propose CLeaD, a supervised contrastive alignment framework that maps WavLM embeddings from English and Mandarin into a shared clinical space, without parallel data or target-language fine-tuning. Evaluating 52 Mandarin speakers, contrastive alignment modestly outperforms the baseline (F1: 0.640 vs. 0.622) under leave-one-speaker-out evaluation. It also improves depressed-class recall at intermediate layers (7-8), though the small test set limits generalizability. Two findings remain robust: model scaling degrades cross-lingual performance while improving monolingual English, and speaker identity leakage artificially inflated previously reported Mandarin F1 scores to 0.954, an artifact we reproduce and quantify.
Original Article
View Cached Full Text

Cached at: 07/08/26, 06:48 AM

Paper page - Layer-wise Cross-Lingual Depression Detection from Speech: Analysis with Contrastive Alignment

Source: https://huggingface.co/papers/2607.02920

Abstract

A supervised contrastive alignment framework maps WavLM embeddings from English and Mandarin into a shared clinical space for depression detection, addressing cross-lingual generalization challenges and revealing performance artifacts caused by speaker identity leakage.

Significant disparities exist in the diagnosis and clinical presentation of depression across different linguistic populations. Speech-based depression detection performs well monolingually, butcross-lingual generalizationremains an open challenge. A key reason is that prior work uses segment-level random splits without speaker grouping, leading to identity leakage that inflates reported metrics. We propose CLeaD, asupervised contrastive alignmentframework that mapsWavLM embeddingsfrom English andMandarininto a shared clinical space, without parallel data or target-language fine-tuning. Evaluating 52Mandarinspeakers, contrastive alignment modestly outperforms the baseline (F1: 0.640 vs. 0.622) underleave-one-speaker-out evaluation. It also improves depressed-class recall at intermediate layers (7-8), though the small test set limits generalizability. Two findings remain robust: model scaling degrades cross-lingual performance while improvingmonolingual English, andspeaker identity leakageartificially inflated previously reportedMandarinF1 scores to 0.954, an artifact we reproduce and quantify.

View arXiv pageView PDFAdd to collection

Get this paper in your agent:

hf papers read 2607\.02920

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2607.02920 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2607.02920 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2607.02920 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Interpretable Depression Detection from Social Media Text Using LLM-Derived Embeddings

arXiv cs.CL

This paper investigates the use of large language models (LLMs) and supervised classifiers for depression detection from social media text, proposing a prompt-based embedding method that enhances interpretability. Experiments on multiple datasets show that zero-shot LLMs perform well for binary classification but struggle with fine-grained severity, while supervised models on LLM summary embeddings achieve more consistent performance across multi-class and ordinal tasks.

Convex Low-resource Accent-Robust Language Detection in Speech Recognition

Hugging Face Daily Papers

This paper introduces CLD, a lightweight convex optimization-based language detection head for ASR that achieves 97-98% accuracy with under 100 training samples while reducing compute costs by 13x, addressing accent and dialect robustness across 5 languages and 24 sub-dialects.