Environmental Sound Classification Using Feature Fusion of MFCCs, Mel-spectrogram, and Chroma

Environmental sound recognition enables automated systems to interpret and respond to diverse acoustic environments, enhancing applications in safety, monitoring, and smart technologies. However, recognizing environmental sounds is challenging due to their diverse types and complex acoustic properties, which encompass a wide range of natural and human-made sounds. To capture the subtleties in different sounds, a robust feature representation is crucial. In this study, we propose a feature fusion-based model for environmental sound classification. For an effective representation of the sounds, we combined three features: MFCCs, mel spectrogram, and chroma STFT. We experimented with several classification algorithms: SVM, LSTM, and BiLSTM. To assess the performance of our proposed model, we utilized the UrbanSound8k dataset. Our study suggests that integrating multiple features can significantly enhance the results compared to relying on individual features. The BiLSTM algorithm enabled our proposed model to achieve a remarkable accuracy of 93.81% on the UrbanSound8k dataset.