Improving analysis of Bengali speech with convolutional neural networks

Abstract

Speech classification is a crucial task in various applications, such as speech recognition and speech analysis. CNNs have shown remarkable success in speech classification tasks for different languages. This research paper aims to enhance Bengali speech classification performance by applying CNNs to a large-scale Bengali speech dataset. The proposed approach includes preprocessing techniques, CNN architecture design, and training strategies specifically tailored for Bengali speech data. Experimental results demonstrate significant improvements in Bengali speech classification accuracy, paving the way for enhanced speech-related applications in the Bengali language. Training strategies are carefully devised, selecting optimization algorithms, learning rate schedules, and batch sizes to maximize classification accuracy of 95.99%. Extensive experiments on the prepared Bengali speech dataset collected from several Facebook posts demonstrate significant improvements in classification accuracy compared to existing methods. The proposed CNN model achieves high accuracy indicating its efficacy in accurately classifying Bengali speech samples. This research contributes to advancing Bengali speech classification using CNNs and showcases the potential of CNNs in processing and analyzing Bengali speech data. This research paper concludes with a thorough investigation of improving Bengali speech categorization using convolutional neural networks. The proposed approach, including dataset preparation, tailored CNN architecture, and optimized training strategies, leads to significant improvements in Bengali speech classification accuracy. The findings highlight the potential impact of CNNs in advancing speech-related applications in the Bengali language.