Development of Intelligent Music Genre Classification Using Stacked LSTM Networks

Classifying music by genre is an essential aspect of modern multimedia retrieval, recommendation systems, and audio analysis. In this study, we propose an intelligent classifier based on a Stacked Long Short-Term Memory (LSTM) network for the automatic identification of music genres from raw audio data. The model leverages the sequential nature of audio features, such as Mel-frequency cepstral coefficients (MFCCs), to learn temporal dependencies and genre-specific patterns. The suggested architecture employs multiple layers of LSTMs to hierarchically capture short- and long-term temporal dynamics in musical signals. A pre-processing pipeline extracts MFCC features from audio clips, which are subsequently passed into the stacked LSTM layers followed by a dense classification layer. The system was trained and evaluated using benchmark datasets such as GTZAN and Free Music Archive (FMA), demonstrating high classification accuracy across diverse genres like classical, jazz, rock, pop, and hip-hop. Experimental results demonstrate the proposed model has achieved the accuracy of 99.7%, precision of 99.6%, recall of 99.2% and F1 score of 99.4%. The proposed model achieved enhanced performance relative to conventional machine learning methods and also outperformed baseline RNNs and shallow LSTMs. Our results highlight the effectiveness of deep recurrent architectures in capturing genre-relevant audio features, making the proposed stacked LSTM-based classifier a promising tool for intelligent applications of music genre classification in real-life scenarios.