In my master's thesis, I participated in a work which investigated the cognitive status of several French and Italian researcher working at the Antarctic Concordia research station. My main goal was to detect the onset of depression based on voice recordings provided by the researchers.
The first step was to collect and summarize the related literature. Afterwards, I segmented the voice recordings at the phonetic level than from these I calculated the voice parameters. I calculated feature vectors from these with mathematical methods. I feed the feature vectors to a Support Vector Regression model developed in the Laboratory of Speech Acoustics. The model gave the predicted Beck Depression Index (BDI) values for each and every sound file.
I plotted the estimated BDI data timewise showing the change in severity of the possible depression in case of every individual researcher. Finally, I compared the estimated BDI results with the mood state data from questionnaires, which the participant researchers filled at the time they made the voice recordings.
The BDI scores as well as the mood state data varied in a close range, therefore, I cannot detect any tendency between them. The estimated BDI results did not permanently go over the value that indicates mild depression. The linkage between the two scales needs further investigation.