Obstructive Sleep Apnea (OSA) is a prevalent sleep disorder characterized by recurrent episodes of upper airway collapse during sleep, leading to significant health consequences. Polysomnography (PSG) remains the gold standard for Obstructive Sleep Apnea diagnosis but is expensive and resource-intensive. This study aims to develop an efficient OSA detection algorithm for classifying OSA snore, normal snore, and other non-snore sound audios. For each audio segment, the time-frequency coefficients are calculated using the discrete wavelet transform (DWT) technique. These extracted features are then fed into a customized Convolution Neural Network (CNN) model to learn discriminative patterns between simple snoring, apneic snoring, and other sleeping sounds. According to the practical analysis, the proposed model has an accuracy of 97.77%, a sensitivity of 98.02%, a precision of 97.87%, and a F1-score of 97.92%. These results demonstrate that the proposed DWT-CNN method achieves superior accuracy compared to existing methods for OSA detection from snoring audio data. This approach has the potential to provide a cost-effective, non-invasive, and automated tool for OSA screening, improving accessibility and early diagnosis of this critical sleep disorder.