RANCANG BANGUN ALAT PERINGATAN DINI BERBASIS IOT DAN MACHINE LEARNING UNTUK PENYANDANG TUNARUNGU DENGAN DETEKSI SUARA KLAKSON KENDARAAN

This study developed an early warning device based on artificial intelligence (AI) and the Internet of Things (IoT) to detect horn sounds in real-time for hearing-impaired individuals. Audio features were extracted using Mel-Frequency Cepstral Coefficients (MFCC) with 40 coefficients over 32 frames, using 4,000 samples (2,000 horn sounds and 2,000 non-horn sounds). The classification model was built using a 1D Convolutional Neural Network (CNN) with three convolutional layers, trained for 15 epochs with a batch size of 32. Training results showed accuracy increasing from 81.25% at epoch 5 to 98.55% at epoch 15, with validation accuracy of 99.10% and low loss (0.0115). Evaluation on the test data showed high precision, recall, and F1-score for both classes, indicating stable classification performance. The model was then integrated into an IoT device consisting of a MAX9814 microphone, ESP32 microcontroller, and vibration motor. Testing demonstrated that the system could consistently provide vibration alerts with an average response time of ±2 seconds. Although effective, further development is needed to expand sound variations and improve device durability in field conditions.
Keywords: sound detection, horn, MFCC, 1D CNN, Internet of Things, hearing impairment, early warning system