# Depression Scale Recognition from Audio, Visual and Text Analysis

**Authors:** Shubham Dham, Anirudh Sharma, Abhinav Dhall  
**arXiv:** [1709.05865](https://arxiv.org/abs/1709.05865)  
**Date:** 2017-09-18  
**Summary page:** https://www.summarizepaper.com/en/paper/1709.05865v1/

## Key points

- Depression is a serious mental health disorder that affects individuals worldwide
- Symptoms of depression can include lack of interest in daily activities, low mood, anxiety, frustration, weight loss and self-hatred
- Shubham Dham, Anirudh Sharma and Abhinav Dhall presented their multimodal feature extraction and decision level fusion approach for the Audio Visual Emotion Challenge (AVEC) 2017 during their second year BTech summer internship
- Their work involved processing data from the Distress Analysis Interview Corpus-Wizard of Oz (DAIC-WOZ) database to extract features using techniques such as Gaussian Mixture Model (GMM) clustering and Fisher vector approach on visual data; statistical descriptors on gaze, pose; low-level audio features; head pose; and text features
- The team used Support Vector Machine (SVM) and neural networks to classify both fused and independent features
- Results showed an improvement over the provided baseline on validation data set by 17% on audio features and 24.5% on video features
- This research highlights the potential for machine learning algorithms to aid in detecting depression automatically through multimodal feature extraction from audio, visual, and text analysis

## Summary

Depression is a serious mental health disorder that affects individuals worldwide, impacting not only their emotional state but also their physical and psychological well-being. Symptoms of depression can include a lack of interest in daily activities, feelings of low mood, anxiety, frustration, weight loss, and even self-hatred. In response to the growing demand for automatic detection of depression using machine learning algorithms, Shubham Dham, Anirudh Sharma, and Abhinav Dhall presented their multimodal feature extraction and decision level fusion approach for the Audio Visual Emotion Challenge (AVEC) 2017 during their second year BTech summer internship. Their work involved processing data from the Distress Analysis Interview Corpus-Wizard of Oz (DAIC-WOZ) database to extract features using techniques such as Gaussian Mixture Model (GMM) clustering and Fisher vector approach on visual data; statistical descriptors on gaze, pose; low-level audio features; head pose; and text features. The team then used Support Vector Machine (SVM) and neural networks to classify both fused and independent features. Their results showed an improvement over the provided baseline on validation data set by 17% on audio features and 24.5% on video features. This research highlights the potential for machine learning algorithms to aid in detecting depression automatically through multimodal feature extraction from audio, visual, and text analysis. By improving our ability to detect depression early on through these methods, we may be able to provide more effective interventions for those suffering from this debilitating condition.