Hasil Pencarian  ::  Simpan CSV :: Kembali

"Penelitian ini melakukan studi mengenai sistem identifikasi pembicara berbahasa Indonesia menggunakan SVM. Parameter sistem terdiri atas silence removal, PCA, nilai rata-rata dan varians MFCC. Ujicoba menggunakan data berita berbahasa Indonesia dari televisi dan radio yang disegmen dalam 5, 10, 15 detik dengan jumlah data 26 jam (715 pembicara). Hasil penelitian ini menunjukkan ketepatan pengenalan pembicara sebesar 94-98% untuk kombinasi parameter silence removal dan rata-rata MFCC dengan akurasi terbaik pada segmen waktu 10 detik. Namun dengan bertambahnya jumlah pembicara, ketepatan pengenalan cenderung berkurang. Penelitian ini dapat dikembangkan untuk sistem perolehan informasi data speech berdasarkan siapa yang berbicara dalam suatu sesi data.

---

This research studies speaker identification system for Indonesian speech based on SVM. Parameters of this system are silence removal, PCA, average and varians values of MFCC. The experiments use 26 hours (715 speakers) Indonesian broadcast news from radio and television segmented into 5, 10, 15 seconds. The results achieve 94-98% identification accuracy for combination of parameters silence removal and average of MFCC. The best accuracy comes from 10 seconds time segment. However, the accuracy falls when the number of speakers increases. This study could be used for speech retrieval system based on who speaks in a speech session."

"ABSTRACT

This study describes bispectrum pattern analysis and quantization for identifying speaker in noisy environment. Direct, non-parametric, bispectrum analysis and estimation was performed before quantization and classification process. As for reliable quantization approach this study applied an algorithm of vector quantization method using combined Self Organizing Feature Map (SOFM) and Learning Vector Quantization (LVQ) neural network, to quantize bispectrum of speech data. Since there is no prior knowledge on bispectrum data distribution to determine class information, we used an adaptive codebook generation method, which is a hybrid of SOFM to generate the codebook internally and LVQ algorithm to improve the cluster distribution in the classifier decision. In addition with the SOFM+LVQ algorithm, a nonlinear vector quantization method (NLVQ) is introduced in dealing with a case where there is a low-separability problem of codebook data obtained from one speaker. This new NLVQ technique employs a nonlinear third order hyperbolic tangent function which combines noise suppression effect with a dynamic range limitation, in or-der to transform the bispectrum input data to be used for making the codebook. Nearest-neighbor rule statistical analysis was used to estimate the recognition performance of the system before classification."

"Penelitian ini mengukur efektivitas perangkat lunak Phonexia Speech Intelligence Resolver (SIR) dalam pemeriksaan audio forensik di Puslabfor Bareskrim Polri. Fokus penelitian meliputi sistem identifikasi pembicara (SID2), identifikasi bahasa (LID2), serta evaluasi akurasi menggunakan Likelihood Ratio (LLR). Proses pra-pemrosesan audio dilakukan menggunakan Audacity, dengan penerapan teknik seperti noise reduction, equalization, compression, enhancement, dan trimming untuk meningkatkan kualitas rekaman.Hasil penelitian menunjukkan bahwa pra-pemrosesan audio memberikan kontribusi signifikan terhadap peningkatan nilai LLR dan kualitas rekaman, yang berdampak langsung pada akurasi identifikasi. Kombinasi lengkap (NR + EQ + C + EN + TS) menghasilkan kualitas terbaik untuk rekaman dengan gangguan berat, sedangkan kombinasi sederhana (NR + EQ + C) lebih efisien untuk rekaman dengan gangguan moderat. Nilai LLR meningkat secara signifikan setelah rekaman diproses, menunjukkan bahwa sistem dapat bekerja lebih optimal pada rekaman berkualitas tinggi.Penelitian ini menegaskan pentingnya pra-pemrosesan sebagai langkah esensial dalam analisis audio forensik. Temuan ini diharapkan dapat memperkuat keandalan Phonexia SIR dalam mendukung proses investigasi kriminal, terutama dengan penerapan metode pra-pemrosesan yang tepat untuk meningkatkan validitas bukti audio.

---

This study evaluates the effectiveness of the Phonexia Speech Intelligence Resolver (SIR) software in forensic audio examination at the Puslabfor Bareskrim Polri. The research focuses on the speaker identification system (SID2), language identification (LID2), and accuracy evaluation using the Likelihood Ratio (LLR) method. Audio pre-processing was conducted using Audacity, employing techniques such as noise reduction, equalization, compression, enhancement, and trimming to improve recording quality.

The findings show that audio pre-processing significantly contributes to increasing LLR values and recording quality, directly enhancing identification accuracy. The complete combination (NR + EQ + C + EN + TS) produced the best results for recordings with heavy noise, while the simpler combination (NR + EQ + C) was more efficient for recordings with moderate noise. LLR values significantly improved after processing, demonstrating that the system performs optimally on high-quality recordings.

This study highlights the importance of pre-processing as an essential step in forensic audio analysis. These findings are expected to strengthen the reliability of Phonexia SIR in supporting criminal investigations, particularly through the implementation of appropriate pre-processing methods to enhance the validity of audio evidence."