Audio Enhancement for Gamelan Instrument Recognition using Spectral Subtraction

Artificial intelligence has made significant progress in processing audio, text, and images, but noise remains a major challenge, especially in real-world audio data. This research presents a novel approach to improve audio classification by integrating noise reduction techniques with machine learning models. Focusing on the bonang barung, a traditional Javanese gamelan instrument, the study uses Mel Frequency Cepstral Coefficients (MFCC) and Mel spectrograms to identify the most effective features for classification, and the Multi-Layer Perceptron (MLP) model for the classification task. In addition, the spectral subtraction method is used to reduce noise, which resulted in significant improvements in audio quality, although some artifacts remain. The main contribution of this study is the integration of noise reduction with the MLP model to improve the classification performance. The MLP model successfully classified various bonang barung playing techniques, achieving a classification accuracy of 90% after noise reduction compared to 87.22% with noise, highlighting the importance of preprocessing steps, such as noise reduction. It is also demonstrated that MLP models can be a viable alternative to more complex deep learning models, such as CNN and RNN, for audio classification tasks. Overall, this research provides new insights into the role of noise reduction in audio analysis and offers potential advances in the field of audio classification.