Abstract
This paper presents a novel approach for analyzing Hindi text data, leveraging a comprehensive methodology that integrates feature extraction using the RESNET50 algorithm and a classification model for detecting Hindi speech patterns. The dataset, initially provided in .mp3 or wav formats, undergoes crucial preprocessing steps. An 80-20 data splitting strategy is employed for training and testing, respectively. A notable aspect of our approach is the utilization of the RESNET50 algorithm, primarily recognized for its excellence in image recognition tasks. However, in this context, we adapt it for audio-related objectives. The extracted features serve as inputs to a classifier designed specifically for detecting Hindi speech patterns. One intriguing but somewhat ambiguous step in our methodology involves the conversion of text into an image.