<p>We performed a survey with 181 volunteers who were tasked to listen to 400 musical extracts from four different genres (rock, pop, classical and electronic) and reported the emotions they perceived along with their intensity. The result is a public dataset called Emotify + with 10 different emotions. It can serve as a research tool in behavioural analysis, sentiment analysis, content analysis and automatic music creation. It can also be used for training small-scale supervised models for various machine learning tasks or simply as ground-truth data for evaluating such methods. In this paper, we provide a detailed report of the dataset and perform a statistical analysis to show the connection of emotions with music genres and other factors. Additionally, we present a baseline predictive model that uses audio features to predict the predominant emotions in a song excerpt. We evaluated two classifiers: support vector machine (SVM) and k-nearest neighbor (KNN). The KNN model significantly outperformed SVM across all performance metrics, achieving a high ROC AUC score (0.81 vs. 0.53), suggesting a more reliable classification. The findings reveal KNN as an effective baseline for music emotion classification in the Emotify dataset, particularly given the complexity of a multiclass task.</p>