Year: 2024
Type: article
Abstract: Effectively communicating and understanding can be a challenging task for people that are either deaf or hard of hearing, it involves them to constantly rely on help to adequately fit in, however with assistive technologies they can minimize their everyday problems. This paper contributes as an advancement to one of these techs and addresses to integrate Faster Whisper, a real-time Automatic Speech Recognition (ASR) model, and a deep learning-based speaker recognition system built on ResNet Convolutional Neural Network (CNN) architecture. Noise Augmentation is employed to enhance the capabilities of the model especially to cope with noisy environments before using the Mel-Scale Frequency Cepstral Coefficients (MFCC) with Delta and Double-Delta Coefficients extracted from speech signals to learn spectral and temporal Features. Due to specific model selection with experiments and skip connections the results have shown significant improvements in Performance metrics with high accuracy and low latency, outperforming standalone models. Future research could explore further refinements in model integration along with the applications of this technology in diverse real-world scenarios, paving the way for more comprehensive solutions in assistive communication technologies.
Authors: Daaman Shah, Rishit Saboo, Aiyush Dwivedi, Megh Gajjar
Institution: Silver Oak University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
Funder: DeepMind
OpenAlex ID: W4402536338