Klasifikasi Suara Darurat dengan Fitur Spektrogram untuk Notifikasi Penyandang Tunarungu

Rakhman, Rizky Abrory (2026) Klasifikasi Suara Darurat dengan Fitur Spektrogram untuk Notifikasi Penyandang Tunarungu. Other thesis, Institut Teknologi Sepuluh November.

Text 5022221209-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Lingkungan jalan raya yang bising dan dinamis menimbulkan risiko keselamatan yang tinggi bagi pengendara penyandang tunarungu, terutama dalam mengenali suara sirene kendaraan darurat. Penelitian ini bertujuan untuk merancang dan mengimplementasikan sistem deteksi serta klasifikasi suara darurat (ambulanss, polisi, dan pemadam kebakaran) berbasis edge computing sebagai perangkat bantu notifikasi visual-taktil. Sistem ini menggunakan mikrofon digital USB yang terintegrasi dengan mikrokomputer Raspberry Pi 4B untuk menangkap sinyal akustik di jalan raya. Untuk mengatasi interferensi noise lingkungan, diterapkan kombinasi filter Bandpass Butterworth orde ke-6 dan algoritma Noise Gate berbasis Root Mean Square (RMS). Sinyal audio kemudian diekstraksi menjadi fitur visual Mel-spektrogram melalui Short-Time Fourier Transform (STFT) untuk diklasifikasikan menggunakan model Convolutional Neural Network (CNN). Output klasifikasi ditransmisikan ke gelang eksternal berupa indikator 4 warna LED dan motor getar. Hasil pengujian menunjukkan bahwa model CNN berhasil mencapai akurasi sebesar 96,4% pada data uji. Total latensi sistem yang dibutuhkan untuk proses akuisisi streaming audio hingga inferensi akhir berkisar antara 3367,1 ms hingga 3392,7 ms. Pada pengujian dinamis di jalan raya, sistem bekerja optimal pada rentang kecepatan berkendara 25–30 km/jam dengan jarak deteksi efektif maksimal 5 meter. Namun, performa sistem mengalami penurunan signifikan pada kecepatan di atas 40 km/jam akibat distorsi wind noise ekstrem pada membran mikrofon.
=====================================================================================================================================
The noisy and dynamic highway environment poses a high safety risk for deaf drivers, particularly in recognizing the sirens of emergency vehicles. This research aims to design and implement an edge computing-based emergency sound detection and classification system (ambulances, police, and fire trucks) as a visual-tactile notification assistive device. The system utilizes a USB digital microphone integrated with a Raspberry Pi 4B microcomputer to capture acoustic signals on the road. To mitigate environmental noise interference, a combination of a 6th-order Butterworth Bandpass Filter and a Root Mean Square (RMS)-based Noise Gate algorithm is applied. The audio signal is then extracted into a Mel-spectrogram visual feature via Short-Time Fourier Transform (STFT) to be classified using a Convolutional Neural Network (CNN) model. The classification output is transmitted to an external bracelet equipped with a 4-color LED indicator and a vibration motor. The experimental results demonstrate that the CNN model achieves an accuracy of 96.4% on the test data. The total system latency required from audio streaming acquisition to final inference ranges from 3367.1 ms to 3392.7 ms. In dynamic highway testing, the system operates optimally within a driving speed range of 25–30 km/h with a maximum effective detection distance of 5 meters. However, the system performance drops significantly at speeds above 40 km/h due to extreme wind noise distortion hitting the microphone membrane

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Penyandang Tunarungu, Sirene Darurat, Raspberry Pi 4B, Mel-spektrogram,Convolutional Neural Network (CNN), Hearing-impaired individuals, Emergency Siren, Raspberry Pi 4B, Mel-spectrogram, Convolutional Neural Network (CNN).
Subjects:	Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) Q Science > QC Physics > QC20.7.F67 Fourier transformations T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5102.9 Signal processing.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20201-(S1) Undergraduate Thesis
Depositing User:	Rizky Abrory Rakhman
Date Deposited:	23 Jul 2026 08:58
Last Modified:	23 Jul 2026 08:58
URI:	http://repository.its.ac.id/id/eprint/136622

Actions (login required)

View Item