In recent years, a key chain that responds to whistling sounds with sound and light has been developed and sold. However, it only responds to whistling sounds with a specific pitch. Therefore, users must have the exact pitch to use it effectively. A system that can identify and respond to whistling sounds regardless of the pitch can solve this issue. We attempted to identify environmental sounds, including whistling sounds, using machine learning. In this study, we obtained a recognition rate of approximately 86% by performing training with a convolutional neural network using mel-frequency cepstrum coefficients as features.