Speech-based Psychological Crisis Assessment using LLMs
2026-05-11 • Computation and Language
Computation and LanguageArtificial Intelligence
AI summaryⓘ
The authors developed a computer program to help classify the level of crisis people are experiencing when they call mental health hotlines. They improved how the program understands emotions by adding information about tone of voice and other sounds that aren't words into the text it reads. They also trained the program to explain its reasoning, which helped it become more accurate. This method showed good results in detecting crisis levels, which could support better hotline services.
large language modelcrisis level classificationparalinguisticsemotional cuesspeech transcriptsreasoning chainsdata augmentationmacro F1-scoremental health hotlinesclassification accuracy
Authors
Terumi Chiba, Yang Luo, Ziyun Cui, Yongsheng Tong, Chao Zhang
Abstract
Psychological support hotlines provide critical support for individuals experiencing mental health emergencies, yet current assessments largely rely on human operators whose judgments may vary with professional experience and are constrained by limited staffing resources. This paper proposes a large language model (LLM)-based framework for automated crisis level classification, a key indicator that supports many downstream tasks and improves the overall quality of hotline services. To better capture emotional signals in spoken conversations, we introduce a paralinguistic injection method that inserts identified non-verbal emotional cues into speech transcripts, enabling LLM-based reasoning to incorporate critical acoustic nuances. In addition, we propose a reasoning-enhanced training strategy that trains the model to generate diagnostic reasoning chains as an auxiliary task, which serves as a regulariser to improve classification performance. Combined with data augmentation, our final system achieves a macro F1-score of 0.802 and an accuracy of 0.805 on the three-class classification task under 5-fold cross-validation.