Machine Learning Applied to Speech Emotion Analysis for Depression Recognition

Depression is a psychiatric condition that can have an impact on a person’s emotional and physical health. Untreated depression can make it difficult for a person to go about their everyday lives and trigger a variety of other symptoms. The only available diagnostic techniques today are therapeutic interventions. As a result, this system is suggested to help with clinical management decisions throughout therapy as well as early detection of depression. As machine learning technology advances, more intelligent systems will soon be usable in daily life. Intelligent systems are interactive, operate with little human effort, and rely mostly on voice input.Hence based on the speech input given, a new computational methodology for detecting various emotions and depression is proposed. Audio samples from the DAIC-WOZ dataset and the RAVDESS dataset were used to create the dataset for depression-related data. Speech signal is exposed to a feature extraction process. Speech corpus is classified using long short-term memory (LSTM) algorithm and its performance is compared with conventionally used Support Vector Machine (SVM). Preliminary result indicates whether individual is depressed or not depressed.