Investigation of speech-based language-independent possibilities of depression recognition

The presented research examined whether it is possible to use the same method to create different language-specific models with similar performance, and whether it is possible to implement language-independent recognition of depression. A depression database one in German and one in Hungarian were used to perform the experiments. The x-vector architecture published by Snyder et al. was used for feature extraction, and Support Vector Regression was used to predict the severity of depression. Classification (depressed / healthy) based on regression results was also implemented. Monolingual and multilingual experiments were performed. Based on the results, it can be stated that it is possible to create different language models with similar performance using the same method. Furthermore, it can be stated that it is possible to create a model valid for multiple languages. Research is currently at an early stage. In the future, it is necessary to expand the number of speech databases used in the experiments.