Automatic Sound Alarm Classification Using Deep Learning For the Deaf and Hard of Hearing
Devis Styo Nugroho, Hendra Kusuma, Tri Arief Sardjono · 2022
Hearing is a very important sensory function for humans. Deaf people who fail to recognize signals indicating dangerous situations around them can endanger their lives, such as building fire alarm signals, gas leak alarms, tsunami alarms, and other dangerous alarms. This research proposes a tool system that can recognize and classify alarm sounds automatically. The two deep learning models we propose for the alarm sound classification task are CNN (Convolutional Neural Network) and LSTM (Long Short Term Memory). The models are trained by using the Mel-Spectrogram extracted from the alarm audio dataset. Based on this experiment, CNN gets better accuracy, reaching 98.83% while the accuracy of the LSTM model is 96.66%. Then the best model will be deployed to the Raspberry Pi, which uses a low-cost microphone for real-time applications.