Curating a strongly labelled urban sound dataset for deep neural network training
The success of deep learning relies on massive training data. However, obtaining large-scale labeled data is not easy, which is expensive and time-consuming. Addressing the complexities of urban soundscapes, this research explores the use of real-world audio data from Singapore to enhance urban soun...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project |
| Language: | English |
| Published: |
Nanyang Technological University
2024
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/177273 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | The success of deep learning relies on massive training data. However, obtaining large-scale labeled data is not easy, which is expensive and time-consuming. Addressing the complexities of urban soundscapes, this research explores the use of real-world audio data from Singapore to enhance urban sound classification. The study identifies the shortcomings of existing sound datasets, which often rely on synthetic or controlled environments. But SINGA:PURA Dataset leverages a comprehensive, real-world dataset to more accurately represent the intricate and dynamic nature of urban noise. This study utilizes advanced machine learning techniques, specifically semi-supervised learning methods like pseudo-labeling, to improve the accuracy and reliability of sound classification systems. This approach aims to develop more robust models, making significant contributions to environmental sound analysis in urban areas where managing noise pollution is critically important. |
|---|