Electronic Theses and Dissertation

Kualitas sinyal suara sering menurun akibat keberadaan noise yang mengurangi energi dan mendistorsi karakteristik spektral asli. Penelitian ini bertujuan memformulasikan model optimasi Second Order Cone Programming (SOCP) untuk memaksimalkan energi sinyal suara sambil tetap mempertahankan karakteristik akustik aslinya. Model dibangun menggunakan teknik Epigraph Form dengan membatasi fitur akustik yang meliputi RMS Energy, Zero-Crossing Rate (ZCR), Spectral Centroid, Pitch Mean, dan MFCC pada interval persentil ke-5 hingga ke-95 dari distribusi dataset global guna menjamin solusi yang robust terhadap outlier. Penyelesaian model dilakukan menggunakan algoritma Primal-Dual Interior-Point yang diimplementasikan menggunakan pustaka NumPy pada bahasa pemrograman Python. Pengujian dilakukan pada dataset Valentini Noisy Speech Database yang terdiri dari 4.698 file audio. Hasil optimasi menunjukkan bahwa dari keseluruhan data yang diproses, sebanyak 3.176 file (67,60%) berhasil mencapai solusi optimal dengan peningkatan energi yang signifikan tanpa menyebabkan distorsi spektral. Sementara itu, 1.522 file (32,40%) teridentifikasi sebagai infeasible karena karakteristik sinyalnya yang secara inheren berada di luar batas wajar. Temuan ini membuktikan bahwa kombinasi nilai energi optimal dan status kelayakan fitur akustik berfungsi efektif sebagai kriteria kuantitatif objektif untuk quality control dalam menyeleksi data audio yang memenuhi standar kualitas penyiaran atau rekaman profesional.

Kata Kunci: Second Order Cone Programming (SOCP), Primal-Dual Interior-Point, Energi Sinyal, Fitur Akustik, Optimasi Konveks

Audio signal quality often degrades due to the presence of noise, which reduces signal energy and distorts its original spectral characteristics. This study aims to formulate a Second Order Cone Programming (SOCP) optimization model to maximize audio signal energy while preserving its original acoustic characteristics. The model is constructed using the Epigraph Form technique, constraining acoustic features including RMS Energy, Zero-Crossing Rate (ZCR), Spectral Centroid, Pitch Mean, and MFCC within the 5th to 95th percentile intervals of the global dataset distribution to ensure a solution robust against outliers. The model is solved using the Primal-Dual Interior-Point algorithm, implemented independently using the NumPy library in Python. The model was evaluated on the Valentini Noisy Speech Database, comprising 4,698 audio files. Optimization results demonstrate that out of the total processed data, 3,176 files (67,60%) successfully achieved optimal solutions with significant energy enhancement without causing spectral distortion. Meanwhile, 1,522 files (32,40%) were identified as infeasible due to their signal characteristics inherently falling outside acceptable bounds. These findings demonstrate that the combination of optimal energy values and acoustic feature feasibility status functions effectively as an objective quantitative criterion for quality control in selecting audio data that meets professional broadcasting or recording standards. Keywords: Second Order Cone Programming (SOCP), Primal-Dual Interior-Point, Signal Energy, Acoustic Features, Convex Optimization