Speech Enhancement using Deep Learning

dc.contributor.author	Mazhar, Muhammad Ali
dc.date.accessioned	2023-07-19T09:33:12Z
dc.date.available	2023-07-19T09:33:12Z
dc.date.issued	2020
dc.identifier.other	172428
dc.identifier.uri	http://10.250.8.41:8080/xmlui/handle/123456789/34826
dc.description	Supervisor: Dr. Ahmad Salman	en_US
dc.description.abstract	While recording speech gets degraded due to the transformations involved and due to interfering noise. In this thesis, we aim to do speech enhancement using deep neural networks i.e. Generative adversarial networks (GANs) for 24 different noise scenarios. We have developed a dataset of 30750 waveforms made using 128 speakers from TIMIT corpus and 6 types of noise each in 4 different severities. We have fine-tuned the available SEGAN model on five different combinations of testing and training waveforms from our dataset. These five different models have given promising results on their respective testing sets in the form of the Signal to Noise (SNR) ratio when compared with the available SEGAN model	en_US
dc.language.iso	en	en_US
dc.publisher	School of Electrical Engineering and Computer Science (SEECS), NUST	en_US
dc.title	Speech Enhancement using Deep Learning	en_US
dc.type	Thesis	en_US