Investigation of Cross-lingual Depression Prediction Possibilities Based on Speech Processing

In this study a cross-lingual investigation is presented about prediction possibilities of depression on the base of speech processing. Our examination was performed on three European languages: German, Hungarian and Italian. Those acoustic features were selected, as input vector of the predictor, which correlate with the severity of depression in quasi language-independent way. Several mono and cross-lingual experiments were conducted. The method is even capable of predicting the severity of depression in the case of a language not used during the training of the model. The experiments clearly show that multilingual depression recognition can be achieved, and it should be possible to construct an automated diagnostic tool for detecting depression, or for patient monitoring, in a cross-lingual way.