Application of autoencoders in speaker recognition system in noisy environment

Abstract

One of the most difficult problems for autonomous speakers is speaker detection and identification because it requires clever technology for the creation of cutting-edge functioning systems. Traditional approaches for Speaker identification and recognition are inaccurate, time-consuming, and have low success rates. This work has been carried out to improve Speaker recognition and identification system accuracy while also increasing success rate. In this study, a dataset gathered from 5 speakers, both men and women, have been rationalized for evaluation of data. The gathered data have been utilized using the Data Augmentation approach based on the size of the dataset. The study implemented a system for recognizing and identifying speakers that makes use of deep learning and autoencoders, specially, in noisy environment. Additionally, in order to validate our result and to prove the novelty, the study compared the results with existing speaker recognition systems.

URN:NBN:sciencein.jist.2024.v12.742