Frequency domain speech enhancement using neural network

• Main Author: Lowis, Albert
• Other Authors: Soon Ing Yann
• Format: Final Year Project
• Language: English
• Published: 2014
• Subjects: DRNTU::Engineering::Electrical and electronic engineering::Electronic systems::Signal processing; DRNTU::Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: http://hdl.handle.net/10356/60899
• Institution: Nanyang Technological University

The project is an exploration of the field of Artificial Intelligence, especially Artificial Neural Network. Artificial Neural Network is a form of Artificial Intelligence that is modelled after the nervous system of animals. The Neural Network is able to learn and be trained like an intelligent system to achieve a specific goal. Previous researches have shown that it has high capability of performing pattern recognition as a computational model. In this project, the Artificial Neural Network is applied specifically to remove noise from noisy speech signal. Although, specifically for this project, the Neural Network does not process the speech signal in time domain, instead it processes it in the Fourier Transform domain. The Neural Network is trained to compare frequency spectrums of the noisy speech signal and the clean speech signal. In the end, the Neural Network shows some capability to objectively improve the noisy speech signal; it can increase the Signal to Noise Ratio, but the resultant speech signal might not be subjectively pleasant for the human ear. As a conclusion, the experiments in this project successfully shows that Artificial Neural Network indeed possess the potential to process speech signals to achieve the goal and further improvements to the result can be made if more resources such as higher computing power is available.