Analisis efektivitas model BiGRU dan BiLSTM dalam penerapan speech recognition untuk aplikasi Smart Home berbasis Arduino

Fadillah, Irma Nur (2025) Analisis efektivitas model BiGRU dan BiLSTM dalam penerapan speech recognition untuk aplikasi Smart Home berbasis Arduino. Sarjana thesis, UIN Sunan Gunung Djati Bandung.

[img]
Preview
Text
1_cover.pdf

Download (132kB) | Preview
[img]
Preview
Text
2_abstrak.pdf

Download (367kB) | Preview
[img]
Preview
Text
3_skbebasplagiarism.pdf

Download (371kB) | Preview
[img]
Preview
Text
4_daftarisi.pdf

Download (274kB) | Preview
[img]
Preview
Text
5_bab1.pdf

Download (1MB) | Preview
[img] Text
6_bab2.pdf
Restricted to Registered users only

Download (5MB) | Request a copy
[img] Text
7_bab3.pdf
Restricted to Registered users only

Download (4MB) | Request a copy
[img] Text
8_bab4.pdf
Restricted to Registered users only

Download (11MB) | Request a copy
[img] Text
9_bab5.pdf
Restricted to Registered users only

Download (334kB) | Request a copy
[img] Text
10_daftarpustaka.pdf
Restricted to Registered users only

Download (1MB) | Request a copy
[img] Text
11_lampiran.pdf
Restricted to Repository staff only

Download (5MB) | Request a copy

Abstract

INDONESIA: INDONESIA: Sistem pengendalian perangkat elektronik tanpa kontak langsung terus berkembang untuk memberikan kemudahan bagi pengguna. Salah satu pendekatan yang umum digunakan adalah sistem klasifikasi perintah suara. Penelitian ini bertujuan untuk merancang dan mengimplementasikan sistem pengenalan suara dengan algoritma Bidirectional Gated Recurrent Unit (BiGRU) dan Bidirectional Long-Short Term Memory (BiLSTM). Metodologi penelitian mencakup perekaman dataset suara dengan lima variasi perintah, penambahan data menggunakan teknik augmentasi, ekstraksi fitur menggunakan MelFrequency Cepstral Coefficients (MFCC) dan Chroma, pelatihan model BiGRU dan BiLSTM dengan optimasi parameter, serta pengujian dalam mode non-realtime sebagai evaluasi awal terhadap data baru, dan pengujian realtime dalam kondisi lingkungan nyata. Hasil pengujian menunjukkan bahwa sistem mampu mengenali perintah suara dalam berbagai kondisi lingkungan, dengan akurasi rata-rata sebesar 82% untuk model BiGRU dan 85% untuk model BiLSTM. Model BiGRU menunjukkan performa yang lebih stabil terhadap suara dari responden tidak terlatih, sedangkan model BiLSTM cenderung lebih unggul pada responden terlatih. Dalam pengujian, sistem menghadapi beberapa tantangan penting, salah satunya adalah kesalahan klasifikasi akibat keterbatasan model dalam mengenali variasi artikulasi dan intonasi dari beberapa responden. Penambahan data melalui tujuh teknik augmentasi terbukti menjadi langkah yang efektif dalam meningkatkan performa model, dengan peningkatan akurasi pengujian realtime hingga 23%. ENGLISH: INGGRIS:The development of contactless electronic device control systems continues to progress in order to provide convenience for users. One commonly used approach is a voice command classification system. This study aims to design and implement a speech recognition system using the Bidirectional Gated Recurrent Unit (BiGRU) and Bidirectional Long Short-Term Memory (BiLSTM) algorithms. The research methodology includes recording a voice dataset with five command variations, augmenting the data using data augmentation techniques, extracting features using Mel-Frequency Cepstral Coefficients (MFCC) and Chroma, training BiGRU and BiLSTM models with parameter optimization, and conducting non-realtime testing as a preliminary evaluation on new data, as well as realtime testing in real-world environmental conditions. The experimental results show that the system is capable of recognizing voice commands under various environmental conditions, with an average accuracy of 82% for the BiGRU model and 85% for the BiLSTM model. The BiGRU model demonstrates more stable performance when handling voices from untrained respondents, while the BiLSTM model tends to perform better with trained respondents. During testing, the system faced several notable challenges, including misclassification due to the model’s limited ability to recognize articulation and intonation variations from certain users. Data augmentation using seven techniques proved to be an effective step in improving model performance, with up to 23% increase in realtime testing accuracy.

Item Type: Thesis (Sarjana)
Uncontrolled Keywords: Speech Recognition; Smart Home; Bidirectional GRU; Bidirectional LSTM; MFCC; Chroma; Arduino UNO
Subjects: Special Computer Methods > Artificial Intelligence
Special Computer Methods > Digital Audio
Applied Physics > Control Divices
Divisions: Fakultas Sains dan Teknologi > Program Studi Fisika
Depositing User: Irma Nur Fadillah
Date Deposited: 13 Oct 2025 08:20
Last Modified: 13 Oct 2025 08:20
URI: https://digilib.uinsgd.ac.id/id/eprint/123223

Actions (login required)

View Item View Item