This study reviews the advancements in detecting depression using audio and text through machine learning. In their overview, the researchers analyzed 65 studies between 2018 and 2014. From these studies, the researchers extracted data on the methods used in these studies, datasets, and other performance metrics to evaluate the approaches used for detecting depression from text or speech. In their results, the authors found that recent advances in multimodal systems that use both text and speech perform a lot better than unimodal systems. Transformer based models like BERT and WavLM achieved promising results, while data augmentation helped increase data diversity. On the other hand, issues such as data scarcity remain major challenges. In the discussion, the authors talked about the need for better quality and culturally diverse data sets.