Urban sound tagging

• Main Author: Lim, Cheng Wei
• Other Authors: Gan Woon Seng
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Electrical and electronic engineering::Computer hardware, software and systems
• Online Access: https://hdl.handle.net/10356/158226
• Institution: Nanyang Technological University

Urban Sound Tagging (UST) seeks to determine whether each of 23 noise sources is present or absent in a 10-second noise by an acoustic sensor network. The 23 noise tags are a multi-label classification problem, and they are common noise complaints in the New York City. The main goal of the competition is to write a computer program to determine whether each of the 23 noise tags is present or absent in the recording. The secondary goal is to classify the 23 fine-grained noise tags and 8 coarse-grained tags. It is sometimes difficult for human to differentiate the closely related noise tags without the use of computer program. For instance, small, medium, and large engines are three fine-grained tags from the coarse-grained engine tag. The absence of noise tag is encoded as 0, while the presence of noise tag is encoded as 1. This report will cover the extraction of baseline Python code using the Git Bash and the Anaconda Juypter Notebook. The interpretations of the Python code to determine the hyperparameters and the model structure of the baseline. The outputs produced by the baseline model code in terms of SoftMax values and loss values. Lastly, the future work and the learning outcome. All the Python codes and the Urban Sound Tagging descriptions in this report were taken from the DCASE community website. [1]