Comparative Study of Optimized ANN and SVM Models for Environmental Sound Classification with Practical Web-Based Deployment

DOI

https://www.doi.org/10.59256/indjcst.20260502098

Published In

Volume 5, Issue 2 (2026)

Published Online

May-August 2026

Pages

904-910

Article Type

Original Article

Check for updates

Environmental Sound Classification (ESC) has already become an important research area in artificial intelligence because of its broad scale of applications in Internet of Things (IoT) systems, intelligent surveillance, smart cities, industrial monitoring and healthcare. This study is performing, a comparative analysis between a Support Vector Machine (SVM) and an Artificial Neural Network(ANN) for multiclass classification of environmental sounds using the UrbanSound8K benchmark dataset. A standardized preprocessing pipeline was developed whereby each audio recording was processed using the librosa library to extract 40 Mel-Frequency Cepstral Coefficients (MFCCs) [4], [15], [16]. The mean and standard deviation of the extracted 40 MFCC coefficients was computed to generate an 80-dimensional acoustic feature vector, which was normalized using Standard Scaler prior to training both models [5]. The ANN architecture was optimized using KerasTuner with the random search algorithm [14] and the SVM employed Radial Basis Function (RBF) kernel [12] . The accuracy, precision, recall, F1- score, confusion matrix, and classification report were utilized to evaluate performance. The overall classification accuracy for the optimized ANN was 94.81%, as compared to 94.66% for the SVM. To demonstrate practical applicability, the optimized ANN was deployed as a Flask-based web application named UrbanSoundAI where users can upload WAV audio recordings and obtain real-time environmental sound predictions with waveforms, Mel spectrogram, and MFCC heatmap visualizations. The results confirm an accurate, repeatable, and practical deployability of the proposed method for environmental sound classification. Moreover, it maintains efficient inference performance for real-world acoustic monitoring applications.