Automatic Sound Detection and Notification System using MFCC

Home accident is one of many issues affecting the household's health and safety. In order to prevent this situation, the smart home device with sound detection and automatic notification system need to be developed. In this work, sound detection system with 6 different important sounds, including the sound of people falling down, the sound of people falling downstairs, the sound of fire, the sound of falling furniture, the sound of broken glass, and the sound of saying “help” are considered. The Mel Frequency Cepstral Coefficient (MFCC) is adopted to represent the sound features, and the TensorFlow Lite (TFLite) is used for classification process. It is found that using the MFCC features along with TSFLite, an accuracy of 99.37% is achieved. The proposed device is done by applying the obtained CNN model into the Raspberry Pi 4 with the microphone. The resulting accuracy more than 70% is obtained when testing in a real situation. Additionally, the automatic notification is successfully transmitted to the smartphone via Line application.