Institutional Repository of Vilnius University
Research on audio noise removal algorithms /
Abstract
dc:descriptionBackground noise reduces the quality and intelligibility of speech recordings. This bachelor’s thesis presents the development and evaluation of an offline audio noise reduction system designed to compare classical and neural network-based speech enhancement methods. The system implements spectral subtraction, Wiener filtering, and a U-Net-based neural network. Each method is implemented as an independent processing pipeline and operates on time-frequency representations of speech signals. When clean reference signals are available, the system evaluates the results using objective speech quality and intelligibility metrics, including SNR, segmental SNR, SI-SNR, STOI, and log-spectral distance. The experimental results show that classical methods perform well under stationary noise conditions but are sensitive to noise estimation errors. The neural network-based method demonstrates more stable performance across different noise conditions. The system provides a structured framework for implementing and experimentally evaluating audio noise reduction techniques.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Nebutovaitė, Živilė,
Subjects
dc:subject × 1Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- eng
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD277691150&prefLang=en_US