Abstract:
Speech categorization is essential for applications like speech recognition and emotionanalysis. Despite the proven effectiveness of Convolutional Neural Networks (CNNs) inspeech categorization, limited research has focused on Bengali speech classification. This
study aims to enhance Bengali speech classification using CNNs on a large-scale dataset. Our approach involves customized preprocessing, CNN architecture, and trainingstrategies for Bengali speech data. Experimental results demonstrate a significant
increase in classification accuracy, achieving 95.99%. The dataset, compiled fromvarious Facebook posts, captures the phonetic and tonal diversity of Bengali. Trainingtechniques were carefully selected, optimizing algorithms, learning rates, and batch sizes. My CNN model's superior accuracy highlights its potential for effectively categorizingBengali speech samples. This research underscores the capability of CNNs in advancingBengali speech categorization and supports the development of speech-relatedapplications in the Bengali language. The study concludes by affirming the substantial
improvements in accuracy achieved through the proposed methods, suggesting newopportunities for Bengali speech applications
