An Automated PSO-CFNN Approach for Speaker Classification and Identification Using a Deep Learning Technique
Year: 2024
Type: article
Abstract: Speaker identification plays a crucial role in numerous applications such as security systems, voice-controlled assistants, forensics, and automated customer services systems. The abilities to accurately identify individuals based on their voices enhance security and the user experiences in these applications making speaker identification an essential part of modern technology. Traditional methods often struggle with background noise and require significant pre-processing, which can limit their effectiveness in the real world. To address these challenges, this study introduced a pioneering PSO-CFNN framework designed to authenticate speaker identification tasks effectively. The proposed framework incorporates several advanced techniques to improve accuracy and robustness. Initially, the wavelet denoising technique is applied to eliminate noise interferences in the audio signals, enhancing the quality of the input data. Then, spectrograms are generated from the denoised audio signal and used as inputs for VGGVox architecture, a deep learning model known for its robust feature extraction capabilities in speaker recognition tasks. Following feature extraction, the Particle Swarm Optimization Algorithm is employed to optimize hyperparameters of Convolutional Fuzzy Neural Networks (CFNN). The optimization step ensures CFNN finely tunes to achieve the best possible performance. In the final stage, CFNN architecture is utilized as a classifier to facilitate automatic speech recognition (ASR). The integration of fuzzy logic within CFNN allows for handling ambiguity and uncertainty in data, further enhancing model robustness.
Source: Nanotechnology Perceptions
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4402136296