Subject: Enhancing Speech Quality In Algerian Dialect Through Deep Learning-Based Denoising Methods

Abstract

This thesis addresses the problem of speech noise reduction, which is one of the main challenges in the field of audio signal processing. Noisy speech signals can significantly degrade the performance of many speech-based applications, such as speech recognition, voice communication, and speech enhancement. In this context, this study explores deep learning-based methods to improve the quality of Algerian dialect speech using denoising techniques. A dataset of 1201 Algerian dialect audio recordings was used, and three deep learning models were developed. These models were tested using four types of noise. The results showed that the Denoising U-Net model outperformed the others in terms of speech clarity and overall denoising performance, making it an effective solution.