This study investigates the impact of audio processing techniques on the performance of Convolutional Neural Networks (CNNs) for snore sound detection, a crucial component in diagnosing Obstructive Sleep Apnea (OSA). OSA affects a significant portion of the population and is associated with serious health risks. We evaluate various processing methods, including raw audio, spectrograms, mel spectrograms (MS), and Mel Frequency Cepstral Coefficients (MFCCs), to determine their effect on the accuracy and efficiency of CNN-based detection models. These models are designed for deployment on resource-constrained edge devices, enabling potential real-time applications. Results indicate that MS processing significantly enhances model accuracy, achieving up to 96% compared to 75% with raw data. Crucially, this improvement is achieved without compromising computational efficiency, making the approach suitable for wearable devices. These findings highlight the critical role of processing techniques in optimizing CNN performance for accurate and scalable snore detection, ultimately contributing to the development of more accessible OSA diagnostic tools.