DESIGN OF SOUNDSCAPE PREDICTION SYSTEM IN URBAN AREA BASED ON NOISE PARAMETER AND MEL-FREQUENCY CEPSTRAL COEFFICIENT (MFCC)

• Main Author: Hanif, Miqdad
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/49885
• Institution: Institut Teknologi Bandung

This research is focused on the design of soundscape prediction system in urban area, based on noise parameter and Mel-Frequency Cepstral Coefficient (MFCC). The objective data are the recorded sound, while the subjective data will be obtained by the questionnaires that will be filled by people in 29 urban areas at Bandung with total 69 sound records. The perception prediction system is designed by using machine learning that use two mathematic models, which are Logistic Regression and Artificial Neural Network (ANN). The result will be compared using Receiver Operating Characteristic (ROC) analysis, in order to obtain the evaluation parameter of Area Under Curve (AUC). The two different approaches and input parameters variation are used to determine the best prediction of perception. Five combination of input parameters are used, which are MFCC, Noise Parameters, MFCC with ?MFCC, MFCC with Noise Parameters, and the combination of those three parameters. The results shows that the best prediction system is for noise perceptions (AUC = 0.836) that system is obtained by using ANN based on MFCC parameters. Therefore the worst system prediction is for echo sound (AUC = 0.645), that system is obtained by using ANN based on MFCC and ?MFCC parameters as input.