An MATLAB Framework for Music Signal Emotion Analysis and Recognition

• Published in: 2023 International Conference on Recent Advances in Electrical, Electronics, Ubiquitous Communication, and Computational Intelligence (RAEEUCCI) pp. 1 - 5
• Main Authors: Sumathi, K, Lakshmidevi, Pasupuleti Baskaran, V, Sindhuja, M, Nivetha
• Format: Conference Proceeding
• Language: English
• Published: IEEE 19.04.2023
• Subjects: Bidirectional Long Short-Term Memory; Emotion recognition; Feature extraction; Long short term memory; Mathematical models; Mel frequency cepstral coefficient; Mel-Frequency Cepstral Coefficients; Multiple signal classification; Music Emotion Recognition; Music Information Retrieval; Training
• DOI: 10.1109/RAEEUCCI57140.2023.10134471

Music has always been a part of our lives and serves both societal and personal needs. Music Emotion Recognition (MER) is a subfield of Music Information Retrieval (MIR) that aims to determine the affective content of music applying machine learning and signal processing techniques. It can be difficult to categorize music according to its emotional content, and a number of challenges need to be resolved, including feature extraction, emotion labelling of music snippets, and algorithm selection. The proposed research-based system recognizes five different major music emotions like calm, energizing, heroic, joyful, and sad from a random input test English music audio data. Initially in the training process, the English music datasets to be trained are pre-processed including noise removal and the Mel-Frequency Cepstral Coefficients (MFCC) features are extracted. And then they undergo Bidirectional Long Short-Term Memory (BI-LSTM) classification. Finally, the trained model is created. Further for the testing process, the input test English music audio data undergoes same pre-processing including noise removal and the Mel-Frequency Cepstral Coefficients (MFCC) features are extracted. And they are then classified with the help of Bidirectional Long Short-Term Memory (BI-LSTM) classification with the help of the pre-trained model. The resultant emotion recognized can be viewed through the created MATLAB GUI. These output data are tabulated and analysed to calculate various essential metrics. This system gives an overall accuracy of 98%, making it an effective system.