MUSIC EMOTION RECOGNITION USING DEEP NEURAL NETWORKS AND LIBROSA

Abstract

In order to arrange, search, and suggest music on contemporary platforms, the categorization of musical emotions is crucial. The deep emotional content included in music may not be fully captured by traditional models, which frequently rely on raw audio or textual data. In order to effectively classify musical emotions, we suggest a Convolutional Neural Network (CNN)-based model in conjunction with Librosa for feature extraction. Mel-frequency cepstral coefficients (MFCCs), chroma characteristics, spectral contrast, and tonette representations are among the significant audio features extracted from music signals using Librosa in the suggested method. By preserving timbral, harmonic, and rhythmic elements important to emotion identification, these aspects offer a condensed and useful representation of the audio. From these collected features, hierarchical patterns are subsequently learned using the CNN model. While pooling layers highlight prominent emotional patterns and minimize dimensionality, convolutional layers automatically identify local correlations in the acoustic information. By removing the need for manually created feature combinations, this deep learning approach enables the model to successfully generalize across a variety of music samples. The suggested approach is able to capture intricate emotional links in music by fusing CNNs' patternlearning capabilities with Librosa feature extraction. This method provides a reliable and scalable solution for automated music emotion classification, supporting real-world platform applications including playlist creation, music analytics, and music recommendation.