Enhancement of Speaker Identification System Based on Voice Active Detection Techniques using Machine Learning
Khine Zin Oo, Lwin Nyein Thu, Zaw Htet Aung · 2024
Speaker identification (SI) is the procedure of classifying the identity of people based on their voice and has now become a useful topic for forensic research. However, SI research has several challenges related to the quality and quantity of the used speech signal that significantly impacts its performance. These are voice variability, insufficient data, and background noise. Thus, active voice detection (AVD) is first employed before extracting features to achieve qualitative signals. In this paper, multiple AVD techniques are conducted to obtain an adequate number of enhanced signals, feature extraction is applied to get relevant features, and the SI system based on machine learning is proposed for identifying the speaker. Experiments are conducted on our dataset. The accuracy of the model using the enhanced signal is almost 5% higher than the accuracy of the model using the original signal.