Leveraging Large Language Models for Spontaneous Speech-Based Suicide Risk Detection

Hong Liu; Jiao Fu; Long Guo; Yifan Gao

REVIEW 3 major objections 5 minor 40 references

Leveraging Large Language Models for Spontaneous Speech-Based Suicide Risk Detection

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A voting ensemble that adds LLM-extracted interpretable indicators to acoustic and text models achieves 74 percent accuracy and F1 for adolescent suicide-risk detection.

desk verdict A sensible challenge system with a genuinely interpretable LLM feature-extraction idea, but the headline 0.740 accuracy is a single point estimate from a small test set and is not statistically robust. read the letter →

arxiv 2507.00693 v1 pith:S3NC7RU2 submitted 2025-07-01 cs.SD cs.CLeess.AS

Yifan Gao , Jiao Fu , Long Guo , Hong Liu This is my paper

classification cs.SDcs.CLeess.AS

keywords suicideriskdetectionspontaneousspeechlargelanguagemodelmultimodalfusioninterpretabilityvotingensembleadolescentmentalhealthDeepSeek-R1

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to establish that a multimodal speech-based system can detect current adolescent suicide risk more accurately than the challenge baselines by adding large-language-model features to conventional acoustic and text models. It reports that a weighted voting ensemble of three speech encoders, one text encoder, and five interpretable indicators extracted by DeepSeek-R1 from an emotion-regulation transcript reaches 74 percent accuracy and F1 on the test set, the top result in the SW1 challenge. This matters because speech can be collected outside clinical settings, and the LLM component returns short verbatim quotes that let a clinician see which risk factors triggered a label.

What carries the argument

The load-bearing mechanism is a structured prompt that asks DeepSeek-R1 to mark five binary indicators in each transcript of the Emotional Regulation task, an open-ended answer about managing extreme distress, namely self-harm behavior, pressure, social support, unhealthy outlets, and exercise, and to quote the supporting text for each. A Random Forest classifies those flags, and its score enters a weighted voting ensemble alongside MLP classifiers trained on HuBERT, Wav2Vec2, Whisper, BERT, and RoBERTa embeddings. The prompt's requirement of verbatim translated quotes is what makes the semantic signal interpretable rather than a black-box score.

What would settle it

Replace the five DeepSeek-R1 flags with random binary flags matched to the same base rates, keep everything else fixed, and re-run the voting ensemble; if test accuracy stays near 0.740, the LLM features are not carrying the signal the paper attributes to them.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that interpretable, LLM-extracted semantic indicators, fused by voting with acoustic and textual predictions, supply information that neither modality alone provides. The winning combination weights outputs from three speech encoders, a multilingual text encoder, and a Random Forest over five flags extracted by DeepSeek-R1 from the Emotional Regulation task, and it reaches 0.740 accuracy and 0.740 F1 on the held-out test set, compared with 0.510 and 0.610 accuracy for the two challenge baselines. The LLM indicator pipeline alone reaches 0.652 accuracy and 0.679 F1, outperforming every single acoustic and text model reported. The authors read this as evidence that transparent, semantically grounded features can carry much of the predictive signal in suicide-risk screening.

Load-bearing premise

The whole voting advantage rests on the five DeepSeek-R1 flags being accurate, but the paper reports no human validation or inter-annotator agreement on those flags; if the flags are wrong or driven by the prompt, the gain over baselines could be an accident of noise.

Editorial extensions

If this is right

An audio-only voting ensemble overfits the development set (0.710 dev, 0.540 test), while adding RoBERTa and the LLM flags raises test accuracy to 0.740, so the interpretable text channel is what stabilises generalization.
DeepSeek-R1's five flags extracted from one ER transcript already beat every single acoustic and text model, so a transcript-only LLM pipeline is a viable lightweight screener.
Because the ensemble returns verbatim transcript snippets for each flagged indicator, a clinician can trace the basis of a high-risk decision instead of receiving an opaque score.
The method ranks first in the SW1 challenge, so on this dataset the combination of pretrained speech encoders, a multilingual text encoder, and LLM-based feature extraction is a workable recipe for spontaneous-speech risk screening.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Nothing in the paper pins down how much of the 0.740 comes from the five LLM flags versus the random forest weighting; ablating each indicator and reweighting the vote would be the direct test.
The five indicators were chosen from prior correlational findings, so the same prompting recipe could be extended to other protective factors; a neighbouring experiment would be to compare prompt-derived flags against human-rated labels on the same transcripts.
Because the conclusion explicitly restricts itself to the MINI-KID current-risk label, a longitudinal follow-up would be needed to know whether a 0.740 decision reflects a stable trait or a momentary state; the paper itself flags this boundary.
The transcripts are Chinese with English-quoted evidence, so the system's portability to other languages is an open question that a cross-lingual replication would settle.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Short version: this is a decent challenge write-up. The new idea that earns its keep is the DeepSeek-R1 prompt that extracts five interpretable binary indicators (self-harm, pressure, social support, unhealthy outlets, exercise) from the ER transcripts, complete with short verbatim quotes, and feeds them through a random forest into a voting ensemble with acoustic and text embeddings. The held-out test result (0.740 accuracy and F1, first in SW1) is a concrete external evaluation, and the authors deserve credit for being explicit in the conclusion that the MINI-KID labels measure current risk, not future suicidal behavior.

The prompt design is genuinely thoughtful, grounded in the psychology literature on protective and risk factors, and the example output in Figure 2 shows a real interpretability mechanism. That is the paper's genuine contribution.

The soft spots are statistical, and they are not minor. The test set is on the order of 100 participants. The winning ensemble was chosen as the best of three on dev; the purely acoustic ensemble (1,2,5,6,8,9) went from 0.710 dev to 0.540 test, a 17-point collapse that tells you how noisy this small sample is. The headline 0.740 has no confidence interval, no significance test, and no per-sample predictions, so we cannot tell whether the margin over Baseline-bonus (0.610) is real or within chance. A rough standard error of 6-7 points puts the 13-point gap at about 1.8-2 SEs; after dev-set selection, the effective strength is weaker. On top of that, the five LLM indicators themselves are unvalidated—no manual scoring, no inter-annotator agreement, no error analysis—so the voting advantage attributed to the interpretable branch could in principle be noise that happens to line up with the dev labels.

None of this kills the idea. The ensemble approach is reasonable for a challenge, and the interpretable LLM feature extraction is a useful contribution to a field that badly needs transparency. But the claim of 'substantially surpassing' the baselines is not statistically established by the evidence in the paper. Anyone building on this should treat the 0.740 as a promising point estimate, not a proven result.

The audience is people working on speech-based mental health screening and challenge participants. Worth a serious referee, because the external evaluation and the interpretability angle deserve engagement. A reviewer should ask for error bars or per-sample outputs, and for some validation of the LLM extraction step.

Referee Report

3 major / 5 minor

Summary. The manuscript describes a system for the 1st SpeechWellness Challenge (SW1) that combines acoustic embeddings from Hubert, Wav2Vec2, and Whisper; text embeddings from BERT and XLM-RoBERTa; and five interpretable indicators extracted from emotional-regulation transcripts by DeepSeek-R1. Each branch is classified separately (MLP for embeddings, Random Forest for LLM indicators), and a weighted voting ensemble yields 0.740 accuracy and F1 on the held-out test set, ranked first in the challenge. The authors claim this substantially surpasses the challenge baselines.

Significance. As a challenge-system paper, the work is of moderate significance. Its positive aspects are the external held-out test benchmark, the rule-based prompt design grounded in cited psychology literature, and the explicit limitation statement about the MINI-KID assessment framework. However, the headline result rests on a single point estimate without statistical inference, and the LLM-extracted features lack validation, which limits the strength of the claims. The interpretability angle is useful but does not offset these verification gaps.

major comments (3)

[Table 3 and Section 4.2] The central claim that the third voting combination 'substantially surpasses' the baselines rests on a single point estimate on the held-out test set. The test set contains roughly 100 participants (Section 2), and no confidence interval, bootstrap, or paired significance test is reported. The winning combination (1,2,5,6,9,13,16) was selected as the best of three on the dev set (0.770 dev vs 0.768 for the alternative), and the acoustic-only ensemble (1,2,5,6,8,9) drops from 0.710 dev to 0.540 test, indicating high variance and dev-set overfitting. With an approximate standard error of 4 to 5 points on the test accuracy, the 13-point margin over Baseline-bonus is about two to three standard errors, but after selection among multiple ensembles on dev, the effective significance is weaker. Please provide confidence intervals, bootstrap results, or per-sample test predictions so that the superiority claim can be evaluated properly.
[Section 3.4 and Figure 2] The five DeepSeek-R1 indicators (self-harm, pressure, social support, unhealthy outlets, exercise) are extracted from ER transcripts with no reported manual validation, inter-annotator agreement, or error analysis. Since the voting weight of the DeepSeek-R1 branch was increased and the ensemble was chosen on the dev set, the contribution of these unvalidated indicators to the 0.740 test accuracy cannot be separated from selection bias. Please provide a manual evaluation on a random sample of transcripts with agreement metrics (e.g., Cohen's kappa) or, at minimum, a detailed error analysis showing that the extracted flags are reliable.
[Section 4.1 and 4.2] The paper reports the final test accuracy without any measure of variance, and it does not state the number of test participants used in Table 3. This makes it impossible to assess whether the 0.740 result is statistically distinct from the 0.610 Baseline-bonus. Please report the test set size, the exact voting rule (including how ties are broken), and a significance test between the proposed ensemble and the baselines on the same test partition.

minor comments (5)

[Section 3.3 heading] The word 'Sementic' should be 'Semantic'.
[Table 3 heading] The heading contains a typo: 'V oting-Based' should be 'Voting-Based'.
[Section 2] The dataset section describes only training and development sets; the test set used in Table 3 is never described, including its size and whether it was provided by the challenge organizers.
[Figure 2 prompt text] The prompt instruction says '1{3 examples'; this appears to be a typo for '1-3 examples'.
[Section 4.1] The manuscript does not mention whether the DeepSeek-R1 API is accessed deterministically (e.g., temperature setting), which is relevant for reproducibility of the indicator extraction.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the central result rests on a held-out challenge test set and externally grounded, non-fitted LLM feature extraction.

full rationale

The paper's derivation chain is not circular. The test set is held out by the SW1 challenge organizers ([20]), and the reported 0.740 accuracy is an external benchmark, not a fitted value. The DeepSeek-R1 interpretable indicators are produced by a frozen LLM using prompt definitions that are themselves grounded in cited psychology literature ([27]–[33]); the indicators are not defined in terms of or fitted to the suicide-risk labels. The RF classifier is trained on these features with the provided labels, which is ordinary supervised classification rather than renaming a fitted parameter as a prediction. The voting ensembles in Table 3 are evaluated on the held-out test set, and the choice among ensembles is based on development-set performance, a standard model-selection step that does not reduce the test claim by construction. The paper contains no self-citations by the present authors that carry a load-bearing argument, and no uniqueness theorem is imported. The absence of confidence intervals and manual validation of LLM indicators are legitimate robustness and statistical concerns, but they do not make the derivation circular. Under the hard rule that circularity must be exhibited by an explicit reduction to the paper's own inputs, none is present.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The result rests on the dataset labels, the reliability of the three speech tasks, and the unvalidated LLM binary indicators. No new entities are introduced.

free parameters (4)

Voting combination and DeepSeek-R1 weight = Chosen from dev set; exact numeric weights not specified
The final combination (1,2,5,6,9,13,16) with increased weight for model 16 was selected based on dev performance (Table 3), affecting the reported 0.740 test result.
MLP hyperparameters = 64, 32, dropout 0.2, lr 1e-4, batch 32, patience 50
These affect per-module classifier performance, though they are standard choices and not the main claim.
Segment length and overlap = 30 seconds, 10% overlap
A preprocessing choice that can change acoustic feature quality, but likely not the load-bearing factor.
Five LLM prompt indicators = Self-harm, pressure, social support, unhealthy outlets, exercise
The set of indicators is selected by the authors from the literature; their reliability is not validated against external annotations.

assumptions (4)

domain assumption MINI-KID self-report suicide module provides valid ground-truth labels for current suicide risk.
The paper itself states the MINI-KID scale is the gold standard for current risk but can underreport or misinterpret symptoms, and results should be interpreted within that assessment framework.
domain assumption The three speech tasks (ER, PR, ED) elicit speech that contains sufficient signal for suicide risk classification.
The whole system depends on the challenge design that these short tasks carry discriminative content; no task-level analysis of sufficiency is provided.
domain assumption ASR transcription of Chinese spontaneous speech is accurate enough for semantic and LLM feature extraction.
The paper does not report ASR word error rate or transcript quality, yet BERT, RoBERTa, and DeepSeek-R1 all consume the transcriptions.
domain assumption Pretrained embeddings (Wav2Vec2, Hubert, Whisper, BERT, XLM-R, DeepSeek-R1) transfer to adolescent Chinese speech and suicide risk content.
No fine-tuning to the domain is described; the method relies on generic pretrained representations.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Leveraging Large Language Models for Spontaneous Speech-Based Suicide Risk Detection." pith.science (2026). https://pith.science/paper/S3NC7RU2

@misc{pith2026250700693,
  author       = {Pith},
  title        = {Pith review of: Leveraging Large Language Models for Spontaneous Speech-Based Suicide Risk Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/S3NC7RU2}},
  note         = {Machine review of arXiv:2507.00693}
}

read the original abstract

Early identification of suicide risk is crucial for preventing suicidal behaviors. As a result, the identification and study of patterns and markers related to suicide risk have become a key focus of current research. In this paper, we present the results of our work in the 1st SpeechWellness Challenge (SW1), which aims to explore speech as a non-invasive and easily accessible mental health indicator for identifying adolescents at risk of suicide.Our approach leverages large language model (LLM) as the primary tool for feature extraction, alongside conventional acoustic and semantic features. The proposed method achieves an accuracy of 74\% on the test set, ranking first in the SW1 challenge. These findings demonstrate the potential of LLM-based methods for analyzing speech in the context of suicide risk assessment.

Figures

Figures reproduced from arXiv: 2507.00693 by the authors.

**Figure 2.** An Example of Using Prompts to Extract DeepSeek Suicide Risk Indicators [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

40 extracted references · 33 canonical work pages

[1]

This mental health issue is particularly prominent among young people, as suicide ranks as the third leading cause of death globally for individuals aged 15 to 29 [ 1]

Introduction Suicide is a significant global mental health issue, with over 720,000 deaths annually and the number of people who attempt suicide is much larger than that of those who die by suicide. This mental health issue is particularly prominent among young people, as suicide ranks as the third leading cause of death globally for individuals aged 15 t...

arXiv 2025
[2]

Have you ever experi- enced moments of extreme emotional distress? How do you manage these feelings ?

Dataset As shown in Table 1. The SW1 dataset [ 20] consists of a training set and a development set (dev). The training set contains data from 400 participants, while the development set includes data from 100. The data distribution in both datasets is the same, with 50% of the participants diagnosed as having a suicide risk. The recordings used for the S...

work page
[3]

First, the audio is fed into an acoustic model to extract acoustic features, while an Automatic Speech Recognition (ASR) model transcribes the speech into text

Method As illustrated in Figure 1, we process raw audio in two main steps. First, the audio is fed into an acoustic model to extract acoustic features, while an Automatic Speech Recognition (ASR) model transcribes the speech into text. Next, the transcription is input into a semantic model to extract text embeddings and into an LLM for LLM features. Final...

work page
[4]

Implementation Details Our model was implemented in PyTorch 2.0.1 on an L20 GPU

Experiment 4.1. Implementation Details Our model was implemented in PyTorch 2.0.1 on an L20 GPU. We employed a MLP classifier with two hidden layers (64 and 32 units, respectively) and a dropout rate of 0.2. We trained the model using Adam (learning rate = 1e-4, batch size = 32) for up to 300 epochs, applying early stopping if the validation metric did no...

work page
[5]

my parents say I’m not good enough

Only use the English translations in brackets, ensuring that the bracketed content is in English (short verbatim quotes). 2.If an indicator is fulfilled (1 = yes), provide the bracketed quotes. If an indicator is not fulfilled (0 = no), do not provide any bracketed text. 3.Provide nothing else and no explanation. Do not add any commentary, summary, or alt...

work page
[6]

at risk” or “no risk,

Conclusion In this study, we propose a multimodal suicide risk detection framework for spontaneous speech that integrates acoustic embeddings, text embeddings, and interpretable features extracted by a large language model (LLM). By combining multiple pretrained models and fusion strategies, our method not only captures essential information—such as emoti...

work page
[7]

Acknowledgement This work is partly supported by the Fundamental Research Funds for Central Universities under grants SCU2024D055 and SCU2024D059

work page
[8]

Suicide,

World Health Organization, “Suicide,” 2024, accessed: 2024-08-29. [Online]. Available: https://www.who.int/zh/ news-room/fact-sheets/detail/suicide

work page 2024

Show all 40 references

[9]

Classification of suicide attempts through a machine learning algorithm based on multiple systemic psychiatric scales,

J. Oh, K. Yun, J.-H. Hwang, and J.-H. Chae, “Classification of suicide attempts through a machine learning algorithm based on multiple systemic psychiatric scales,” Frontiers in Psychiatry, vol. 8, p. 192, 2017

2017
[10]

The beck depression inventory-ii (bdi-ii), beck hopelessness scale (bhs), and beck scale for suicide ideation (bss)

D. J. Dozois and R. Covin, “The beck depression inventory-ii (bdi-ii), beck hopelessness scale (bhs), and beck scale for suicide ideation (bss).” 2004

2004
[11]

A review of depression and suicide risk assessment using speech analysis,

N. Cummins, S. Scherer, J. Krajewski, S. Schnieder, J. Epps, and T. F. Quatieri, “A review of depression and suicide risk assessment using speech analysis,” Speech communication, vol. 71, pp. 10–49, 2015

2015
[12]

New methods for assessing rapid changes in suicide risk,

E. D. Ballard, J. R. Gilbert, C. Wusinich, and C. A. Zarate Jr, “New methods for assessing rapid changes in suicide risk,” Frontiers in psychiatry, vol. 12, p. 598434, 2021

2021
[13]

Johar, Emotion, affect and personality in speech: The Bias of language and paralanguage

S. Johar, Emotion, affect and personality in speech: The Bias of language and paralanguage. Springer, 2015

2015
[14]

Acoustic and language analysis of speech for suicidal ideation among us veterans,

A. Belouali, S. Gupta, V . Sourirajan, J. Yu, N. Allen, A. Alaoui, M. A. Dutton, and M. J. Reinhard, “Acoustic and language analysis of speech for suicidal ideation among us veterans,” BioData Mining, vol. 14, no. 1, pp. 1–17, 2021

2021
[15]

Linguistic features of suicidal thoughts and behaviors: A systematic review,

S. Homan, M. Gabi, N. Klee, S. Bachmann, A.-M. Moser, M. Duri’, S. Michel, A.-M. Bertram, A. Maatz, G. Seiler, E. Stark, and B. Kleim, “Linguistic features of suicidal thoughts and behaviors: A systematic review,”Clinical Psy- chology Review, vol. 95, p. 102161, 2022

2022
[16]

Exploring the role of first-person singular pronouns in detecting suicidal ideation: A machine learning analysis of clinical transcripts,

R. Huang, S. Yi, J. Chen, K. Y . Chan, J. W. Y . Chan, N. Y . Chan, S. X. Li, Y . K. Wing, and T. M. H. Li, “Exploring the role of first-person singular pronouns in detecting suicidal ideation: A machine learning analysis of clinical transcripts,” Behavioral Sciences, vol. 14,...

2024
[17]

The indirect effect of per- ceived criticism on suicide ideation and attempts,

C. R. Hagan and T. E. Joiner, “The indirect effect of per- ceived criticism on suicide ideation and attempts,” Archives of Suicide Research, vol. 21, no. 3, pp. 438–454, 2017, pMID: 27487316

2017
[18]

Specific coping behaviors in relation to adolescent depression and suicidal ideation,

“Specific coping behaviors in relation to adolescent depression and suicidal ideation,” Journal of Adolescence, vol. 34, no. 5, pp. 1077–1085, 2011

2011
[19]

Latent suicide risk detection on microblog via suicide- oriented word embeddings and layered attention,

L. Cao, H. Zhang, L. Feng, Z. Wei, X. Wang, N. Li, and X. He, “Latent suicide risk detection on microblog via suicide- oriented word embeddings and layered attention,” arXiv preprint arXiv:1910.12038, 2019

1910 arXiv
[20]

Validating machine learning algorithms for twitter data against established measures of suicidality,

S. R. Braithwaite, C. Giraud-Carrier, J. West, M. D. Barnes, and C. L. Hanson, “Validating machine learning algorithms for twitter data against established measures of suicidality,” JMIR mental health, vol. 3, no. 2, p. e4822, 2016

2016
[21]

Using large language models to detect depression from user-generated diary text data as a novel approach in digital mental health screening: Instrument validation study,

D. Shin, H. Kim, S. Lee, Y . Cho, and W. Jung, “Using large language models to detect depression from user-generated diary text data as a novel approach in digital mental health screening: Instrument validation study,” Journal of Medical Internet Research, vol. 26, p. e54617, 2024

2024
[22]

Linguistic features ex- tracted by gpt-4 improve alzheimer’s disease detection based on spontaneous speech,

J. Heitz, G. Schneider, and N. Langer, “Linguistic features ex- tracted by gpt-4 improve alzheimer’s disease detection based on spontaneous speech,” arXiv preprint arXiv:2412.15772 , 2024

2024 arXiv
[23]

Suicidal intention de- tection in tweets using bert-based transformers,

G. Ananthakrishnan, A. K. Jayaraman, T. E. Trueman, S. Mi- tra, A. Abinesh, and A. Murugappan, “Suicidal intention de- tection in tweets using bert-based transformers,” in 2022 In- ternational Conference on Computing, Communication, and Intelligent Systems (ICCCIS). IEEE, 2022...

2022
[24]

Detecting potentially harmful and protective suicide-related content on twitter: machine learning approach,

H. Metzler, H. Baginski, T. Niederkrotenthaler, and D. Garcia, “Detecting potentially harmful and protective suicide-related content on twitter: machine learning approach,” Journal of medical internet research, vol. 24, no. 8, p. e34705, 2022

2022
[25]

Spontaneous speech-based suicide risk detection using whisper and large language models,

Z. Cui, C. Lei, W. Wu, Y . Duan, D. Qu, J. Wu, R. Chen, and C. Zhang, “Spontaneous speech-based suicide risk detection using whisper and large language models,” arXiv preprint arXiv:2406.03882, 2024

2024 arXiv
[26]

Multimodal sensing for depression risk detection: integrating audio, video, and text data,

Z. Zhang, S. Zhang, D. Ni, Z. Wei, K. Yang, S. Jin, G. Huang, Z. Liang, L. Zhang, L. Li et al. , “Multimodal sensing for depression risk detection: integrating audio, video, and text data,” Sensors, vol. 24, no. 12, p. 3714, 2024

2024
[27]

The 1st speechwellness challenge: Detecting suicidal risk among adolescents,

W. Wu, Z. Cui, C. Lei, Y . Duan, D. Qu, J. Wu, B. Zhou, R. Chen, and C. Zhang, “The 1st speechwellness challenge: Detecting suicidal risk among adolescents,” 2025. [Online]. Available: https://arxiv.org/abs/2501.06474

2025
[28]

Unsupervised cross-lingual representation learning for speech recognition,

A. Conneau, A. Baevski, R. Collobert, A. Mohamed, and M. Auli, “Unsupervised cross-lingual representation learning for speech recognition,” arXiv preprint arXiv:2006.13979 , 2020

2006 arXiv
[29]

Robust speech recognition via large-scale weak supervision,

A. Radford, J. W. Kim, T. Xu, G. Brockman, C. McLeavey, and I. Sutskever, “Robust speech recognition via large-scale weak supervision,” in International conference on machine learning. PMLR, 2023, pp. 28 492–28 518

2023
[30]

Hubert: Self-supervised speech representation learning by masked prediction of hidden units,

W.-N. Hsu, B. Bolte, Y .-H. H. Tsai, K. Lakhotia, R. Salakhut- dinov, and A. Mohamed, “Hubert: Self-supervised speech representation learning by masked prediction of hidden units,” IEEE/ACM transactions on audio, speech, and language pro- cessing, vol. 29, pp. 3451–3460, 2021

2021
[31]

Bert: Pre-training of deep bidirectional transformers for language understanding,

J. Devlin, “Bert: Pre-training of deep bidirectional transformers for language understanding,” arXiv preprint arXiv:1810.04805, 2018

2018 arXiv
[32]

Unsupervised cross-lingual representation learn- ing at scale,

A. Conneau, “Unsupervised cross-lingual representation learn- ing at scale,” arXiv preprint arXiv:1911.02116, 2019

1911 arXiv
[33]

Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning,

D. Guo, D. Yang, H. Zhang, J. Song, R. Zhang, R. Xu, Q. Zhu, S. Ma, P. Wang, X. Bi et al. , “Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning,”arXiv preprint arXiv:2501.12948, 2025

2025 arXiv
[34]

Social connectedness as a determi- nant of mental health: A scoping review,

P. J. Wickramaratne, T. Yangchen, L. Lepow, B. G. Patra, B. Glicksburg, A. Talati, P. Adekkanattu, E. Ryu, J. M. Bier- nacka, A. Charney et al., “Social connectedness as a determi- nant of mental health: A scoping review,”PloS one, vol. 17, no. 10, p. e0275004, 2022

2022
[35]

Differences in non- suicidal self-injury and suicide attempts in adolescents,

A. M. Brausch and P. M. Gutierrez, “Differences in non- suicidal self-injury and suicide attempts in adolescents,” Jour- nal of youth and adolescence , vol. 39, pp. 233–242, 2010

2010
[36]

Social support as a protective factor in suicide: Findings from two nationally representative samples,

E. M. Kleiman and R. T. Liu, “Social support as a protective factor in suicide: Findings from two nationally representative samples,” Journal of affective disorders, vol. 150, no. 2, pp. 540–545, 2013

2013
[37]

Interaction between physical activity and prob- lematic mobile phone use on suicidality in chinese college students,

Y . Xie, M. Zhu, X. Wu, S. Tao, Y . Yang, T. Li, L. Zou, H. Xu, and F. Tao, “Interaction between physical activity and prob- lematic mobile phone use on suicidality in chinese college students,” BMC psychiatry, vol. 20, pp. 1–7, 2020

2020
[38]

Physical exercise, mental health problems, and suicide attempts in university students,

M. Grasdalsmoen, H. R. Eriksen, K. J. Lønning, and B. Sivert- sen, “Physical exercise, mental health problems, and suicide attempts in university students,” BMC psychiatry, vol. 20, pp. 1–11, 2020

2020
[39]

Coping strategies and suicidality: A cross- sectional study from china,

J. Liang, K. K˜olves, B. Lew, D. de Leo, L. Yuan, M. Abu Talib, and C.-x. Jia, “Coping strategies and suicidality: A cross- sectional study from china,” Frontiers in Psychiatry, vol. 11, 2020

2020
[40]

An examination of the mediating role of maladap- tive emotion regulation strategies in the complex relationship between interpersonal needs and suicidal behavior,

H. Abdollahpour Ranjbar, M. Bakhshesh-Boroujeni, S. Farajpour-Niri, I. Hekmati, M. Habibi Asgarabad, and M. Eskin, “An examination of the mediating role of maladap- tive emotion regulation strategies in the complex relationship between interpersonal needs and suicidal behavior...

2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

This mental health issue is particularly prominent among young people, as suicide ranks as the third leading cause of death globally for individuals aged 15 to 29 [ 1]

Introduction Suicide is a significant global mental health issue, with over 720,000 deaths annually and the number of people who attempt suicide is much larger than that of those who die by suicide. This mental health issue is particularly prominent among young people, as suicide ranks as the third leading cause of death globally for individuals aged 15 t...

arXiv 2025

[2] [2]

Have you ever experi- enced moments of extreme emotional distress? How do you manage these feelings ?

Dataset As shown in Table 1. The SW1 dataset [ 20] consists of a training set and a development set (dev). The training set contains data from 400 participants, while the development set includes data from 100. The data distribution in both datasets is the same, with 50% of the participants diagnosed as having a suicide risk. The recordings used for the S...

work page

[3] [3]

First, the audio is fed into an acoustic model to extract acoustic features, while an Automatic Speech Recognition (ASR) model transcribes the speech into text

Method As illustrated in Figure 1, we process raw audio in two main steps. First, the audio is fed into an acoustic model to extract acoustic features, while an Automatic Speech Recognition (ASR) model transcribes the speech into text. Next, the transcription is input into a semantic model to extract text embeddings and into an LLM for LLM features. Final...

work page

[4] [4]

Implementation Details Our model was implemented in PyTorch 2.0.1 on an L20 GPU

Experiment 4.1. Implementation Details Our model was implemented in PyTorch 2.0.1 on an L20 GPU. We employed a MLP classifier with two hidden layers (64 and 32 units, respectively) and a dropout rate of 0.2. We trained the model using Adam (learning rate = 1e-4, batch size = 32) for up to 300 epochs, applying early stopping if the validation metric did no...

work page

[5] [5]

my parents say I’m not good enough

Only use the English translations in brackets, ensuring that the bracketed content is in English (short verbatim quotes). 2.If an indicator is fulfilled (1 = yes), provide the bracketed quotes. If an indicator is not fulfilled (0 = no), do not provide any bracketed text. 3.Provide nothing else and no explanation. Do not add any commentary, summary, or alt...

work page

[6] [6]

at risk” or “no risk,

Conclusion In this study, we propose a multimodal suicide risk detection framework for spontaneous speech that integrates acoustic embeddings, text embeddings, and interpretable features extracted by a large language model (LLM). By combining multiple pretrained models and fusion strategies, our method not only captures essential information—such as emoti...

work page

[7] [7]

Acknowledgement This work is partly supported by the Fundamental Research Funds for Central Universities under grants SCU2024D055 and SCU2024D059

work page

[8] [8]

Suicide,

World Health Organization, “Suicide,” 2024, accessed: 2024-08-29. [Online]. Available: https://www.who.int/zh/ news-room/fact-sheets/detail/suicide

work page 2024