Design and Implementation of Sound-based IoT Systems for Environment Control and monitoring

Please enable javascript in your browser.

Samimifar, Milad | 2024

0 Viewed

Type of Document: M.Sc. Thesis
Language: Farsi
Document No: 57162 (05)
University: Sharif University of Technology
Department: Electrical Engineering
Advisor(s): Gholampour, Iman
Abstract:
The processing of audio signals can be very widely used in the Internet of Things because by using audio signals, valuable information can be obtained from the activities of the environment. Environmental sound processing and investigation can be used in applications such as industrial troubleshooting security systems, etc. For this, it is necessary to build a sound recognition and classification system. In this thesis, we have evaluated the features obtained from feature engineering and feature learning and the use of various machine learning methods for the audio event classification system. To test and evaluate the system, known data in this field have been used. Finally, by proposing an optimal neural network for voice classification, we achieved 90% accuracy on ESC-50 data, which compared to other similar methods on this data and human accuracy, we have seen a few percent improvement. In the following, to compress the models with the aim of easy transfer and sufficient execution speed, using optimization methods, we see a seventy-five percent reduction in the volume of the obtained models
Keywords:
Internet of Things ; Machine Learning ; Deep Learning ; Sound Events ; Microcontroller