SpeechT-RAG: Reliable Depression Detection in LLMs with Retrieval-Augmented Generation Using Speech Timing Information
Xiangyu Zhang, Hexin Liu, Qiquan Zhang, Beena Ahmed, Julien Epps, The University of New South Wales, Nanyang Technological University · 2025
Read SpeechT-RAG: Reliable Depression Detection in LLMs with Retrieval-Augmented Generation Using Speech Timing Information by Xiangyu Zhang, Hexin Liu, Qiquan Zhang, Beena Ahmed, Julien Epps, The University of New South Wales, Nanyang Technological University (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-10950.txt.
Description
Xiangyu Zhang, Hexin Liu, Qiquan Zhang, Beena Ahmed, and Julien Epps detect depression, in a 2025 preprint, from the timing of speech rather than its words alone. Language models asked to judge depression from text do only fairly. Adding a standard retrieval step barely helps. The missing signal, the authors argue, sits in how speech unfolds in time: pauses, pace, and rhythm. They first compare those timing patterns between healthy speakers and speakers with depression. Clear differences show up.
Their system, SpeechT-RAG, retrieves reference cases by timing traits instead of text. The same traits drive a confidence score, so the system knows when to trust itself. It matched fine-tuned models on accuracy with no extra training. Unlike them, it also reports how sure it is. The authors frame that pairing, accuracy plus honesty about doubt, as what mental health tools require before real use.
Catalog record
| CREATOR | Xiangyu Zhang, Hexin Liu, Qiquan Zhang, Beena Ahmed, Julien Epps, The University of New South Wales, Nanyang Technological University |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 36 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-10950