Sound-based insect categorization has emerged as a practical method for monitoring biodiversity and ecological health. In order to classify insect species, this research presents a deep learning framework that combines the Short-Time Fourier Transform (STFT), log mel spectrograms, and Convolutional Neural Networks (CNNs). To train and assess the model, 24 insect species that belong to the family Orthoptera, have been used. To transform the time-domain signals into frequency-domain representations, the suggested method starts with preprocessing the raw audio data using STFT. From the STFT outputs, log mel spectrograms are subsequently produced, which provide a more perceptually meaningful depiction of the insect sounds. These spectrograms are what the CNN model uses as input features. The spectrograms intricate patterns and properties are efficiently captured by the CNN architecture, allowing for precise classification of insect species. The model's resilience is demonstrated by the performance evaluation, which yields a classification accuracy of 98.66%