The study of attention mechanism in sound classification
In modern society today, high rise building and motor vehicles exist all around us. The robust sound events detected within the environment can be heard clearly by us. However, this might not be the case for machines. I.e. Robots. Sounds captured by machines, are often filled with interference from...
Saved in:
Main Author: | |
---|---|
Other Authors: | |
Format: | Final Year Project |
Language: | English |
Published: |
Nanyang Technological University
2021
|
Subjects: | |
Online Access: | https://hdl.handle.net/10356/149298 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Institution: | Nanyang Technological University |
Language: | English |
Summary: | In modern society today, high rise building and motor vehicles exist all around us. The robust sound events detected within the environment can be heard clearly by us. However, this might not be the case for machines. I.e. Robots. Sounds captured by machines, are often filled with interference from the environment. As a result, when specific sounds such as breaking of glass or even car honks are detected, machines are unable to identify the sound event accurately.
This created the idea of developing an automated sound classifier using the deep-learning model. By leveraging on the attention mechanism used in transformers in combination with CNN, the new system allows multiple sound events to be attended at the same time which promotes a parallel computation approach as compared to other neural networks such as the Recurrent Neural Network (RNN) and Long-short term memory (LSTM) which suffers from a series computation approach. Overall, The performances of the network models are evaluated based on different parameters to get a comparison of which network models is best suited for time-series dataset. |
---|