The detection of mental disorders through Natural Language Processing (NLP) has gained significant attention, particularly in the field of depression detection. Despite the availability of various corpora for English, Spanish-language resources remain limited. To address this gap, this study presents a synthetic corpus based on Plutchik’s emotion model, designed to enhance depression detection in Spanish-language social media messages. This work compares three different corpora: the Twitter Sentiment analysis corpus, the LiSSS corpus, and the created synthetic corpus. Each dataset is used to train emotion-detection models, which then generate emotion vectors as input for a multilayer perceptron neural network to classify users as depressed or non-depressed. The results demonstrate the effectiveness of emotion-based features for depression detection and highlight the potential of synthetic corpora. The comparison shows that the synthetic corpus is statistically similar to a corpus generated and labeled by humans. Moreover, the results of the model based on the synthetic corpus perform as well as or better than other corpora labeled by humans.