ABSTRACT
Vocal emotions are basic to expressing and understanding thoughts and the low toned vocal emotion expression is a major deficit in individuals with depression. The primary objective of this paper is to propose a federated learning (FL)-based classification model to identify depression in individuals through audio. The current scenario of artificial intelligence (AI) focuses on collaborative training of deep learning (DL) models without losing data privacy. So, in the methodology of this paper, a collaborative and privacy preserved approach has been developed using FL for training deep learning models. The long-short short-term memory (LSTM) and bidirectional-long-short term memory (B-LSTM)-based deep learning models will be trained on a collected dataset in the federated learning ecosystem. As a result, the implemented models will be comparatively analyzed on base DL structure as well as the FL ecosystem. The purpose of the investigation is to compare the impact of FL architecture implementations on benchmark models. In conclusion, the most effective examined strategy will be considered for future research objectives.
