Translation. Region: Russian Federation –
Source: Novosibirsk State University –
An important disclaimer is at the bottom of this article.
A research group from NSU, which includes scientists and students from the Psychology department Institute of Medicine and Medical Technologies (IMMT) NSU, developed an automated approach to assessing depression based on the acoustic characteristics of human speech. The project was supported by the program "Priority 2030".
Today, depression is one of the most common mental disorders. According to WHO estimates for 2025 Depression affects approximately 332 million people worldwide. Depression can occur as a standalone disorder or coexist with other illnesses, including physical ones. The situation is complicated by the fact that depression is often masked by physical complaints: patients experience vague aches and pains (for example, frequent headaches), heart problems, digestive problems, and a general deterioration in well-being, while the symptoms remain ambiguous, and it is impossible to determine the physical cause. In this situation, general practitioners often lack the time and expertise to conduct an in-depth diagnostic evaluation and make an accurate diagnosis.
"Analysis of objective indicators of depression can help reduce the workload of doctors and ensure accurate, high-quality, and timely diagnosis. Voice analysis can serve as one such indicator. It's worth noting that diagnosing depression using acoustic voice characteristics offers several advantages over traditional methods based on self-reporting and interviews, primarily because it completely eliminates the factor of social desirability: it's an objective indicator that reflects a person's condition, while a person cannot voluntarily control their voice to conceal symptoms of depression (or, conversely, aggravate them)," explained Marina Zlobina, PhD in Psychology, Senior Lecturer in the Department of Personality Psychology at the Institute of Mathematical and Mathematical Technologies (IMMT) of Novosibirsk State University, and the project's director.
A considerable number of studies have already been published abroad on diagnosing depression based on acoustic voice characteristics, including using neural network approaches. However, there is no data yet on the practical application of such technology. In Russia, such solutions are only just beginning to emerge: for example, technologies for assessing a person's condition based on voice parameters are being developed as part of research into human functional states in spaceflight conditions. However, these technologies do not address the issue of diagnosing anxiety and depression.
As the project's authors note, speech is a natural biomarker of mental state. Even a short excerpt contains valuable information about vocal energy, which is subject to change in depression and anxiety. Based on over 90 interviews, the researchers trained a neural network model that classifies speech into four levels of depression severity—from no symptoms to severe. The PHQ-9 questionnaire was used to assess the severity of depressive symptoms.
— В основу разработки легла современная архитектура wav2veс, которая позволяет извлекать векторные акустические характеристики голоса. Обученная модель демонстрирует высокую точность, которая сопоставима с результатами ведущих зарубежных исследований: точность оценивалась на основе показателя F1 — гармоническое среднее между точностью (precision) и полнотой (recall), F1 достиг значения >0.94. For practical use of the technology, a prototype NeuroVoice GUI application was created, implemented using the PyQt5 framework. The interface supports the full data management cycle—from uploading or recording audio to visualizing results and exporting recordings. The prototype allows both uploading existing recordings and conducting on-the-fly evaluations, explained Alexander Fedorov, PhD in Psychology, Associate Professor, and Head of the Department of Clinical Psychology at the Institute of Medical and Mathematical Technologies (IMMT) of Novosibirsk State University.
Work on the project was carried out over a period of four months – from September to December 2025. The team included Alexander Fedorov, PhD in Psychology, Associate Professor, Head of the Department of Clinical Psychology at IMMT; Marina Zlobina, PhD in Psychology, Senior Lecturer in the Department of Personality Psychology at IMMT; Kirill Kirilenkov, a graduate of the Psychology program at IMMT NSU; and Psychology students Daria Farkova (4th year) and Anastasia Glazunova (3rd year).
"It's important to note that this technology is not a replacement for a specialist psychologist or psychiatrist. However, it can be effectively used by general practitioners to identify comorbid depression associated with somatic illnesses, as well as masked depression, which often manifests as physical complaints, pain, and malaise that cannot be explained by a somatic illness," added Marina Zlobina.
The project is quite promising: plans call for expanding the dataset and using it to create a database of interviews with Russian-speaking subjects, similar to the English-language DAIC-WOZ (a multimodal corpus consisting of audio and video recordings, as well as transcribed interview text). Models will then be trained on the expanded dataset, integrated into the final application, and subsequently tested.
"In the future, the developed technology could also be used to diagnose other mental disorders (provided there is additional data available to further train the model). Furthermore, it is possible to integrate additional modalities (for example, facial expression analysis from video recordings)," concluded Marina Zlobina.
Please note: This information is raw content obtained directly from the source. It represents an accurate account of the source's assertions and does not necessarily reflect the position of MIL-OSI or its clients.