ABSTRAKSI: Saat ini, penggunaan audio seperti aplikasi media yang real-time sangatlah dibutuhkan, terutama pada media aplikasi yang menggunakan proses diferensiasi pada data audio, seperti pengkodean yang berbasis konten dan kompresi audio maupun penyetaraan antara speech dan music secara otomatis. Oleh karena itu, diperlukan suatu algoritma yang efisien untuk melakukan segmentasi sinyal audio menjadi speech signal ataupun music signal. Dalam tugas akhir ini, digunakan suatu pendekatan untuk mendeteksi batasan-batasan musik dan mengklasifikasikan speech / music dengan menggunakan suatu algoritma yang dinamakan segmental continuous dynamic programming atau disingkat dengan Segmental CDP.
Algorima Segmental CDP dapat mengidentifikasi lokasi dari masing-masing bagian musik dan batasan-batasannya berdasarkan berbagai kesamaan segmen dan informasi lokasinya.
Ekstraksi ciri yang digunakan dalam domain waktu diberikan dua pilihan, yaitu : ZCR (Zero Crossing Rate) dan Energi Bit, sedangkan MFCC merupakan ekstraksi ciri dalam domain frekuensi. Pemisahaan sinyal campuran berhasil dilakukan dengan menggunakan threshold dari ciri tersebut. Sinyal audio dikategorikan sebagai speech signal jika nilai moving average energy bit ≤ nilai maksimum moving average energy bit speech, nilai moving average ZCR ≥ nilai minimum moving average ZCR speech, dan nilai moving average MFCC ≤ nilai maksimum moving average MFCC speech. Sinyal audio dikategorikan sebagai music signal jika, nilai moving average energy bit ≥ nilai minimum moving average energy bit music, nilai moving average ZCR ≤ nilai maksimum moving average ZCR music,dan nilai moving average MFCC ≥ nilai minimum moving average MFCC music.
Kata Kunci : segmental CDP, speech, music, segmentasi, klasifikasiABSTRACT: Nowadays, the use of audio as the application of real-time media is desperately needed, especially in media applications using the differentiation process on audio data, such as content based encoding and audio compression and equalization between speech and music automatically. Therefore, it is required an efficient algorithm to segment the audio signal into speech signal or music signal. In this final project, we use an approach for detecting music boundaries and classify speech / music by using an algorithm called segmental continuous dynamic programming or shortened by Segmental CDP.
Segmental CDP algorithm can be used to identify the location of each piece of music and their limits based on various similarity segment and location information.
Feature extraction in the time domain is given two options, namely: ZCR (Zero Crossing Rate) and Bit Energy, while the MFCC feature extraction is in the frequency domain. Separation in mixed signals successfully performed using the threshold of the traits. The audio signal is categorized as a speech signal if the value of moving average energy bit ≤ maximum value of the moving average bit energy speech, a moving average value of ZCR ≥ minimum value of ZCR speech moving average, moving average MFCC and value ≤ maximum value of moving average MFCC speech. The audio signal is categorized as a music signal if, the value of moving average energy bit ≥ minimum value of moving average energy bit music, moving average ZCR value ≤ the maximum value of moving average ZCR music, and moving average MFCC value ≤ minimum value of moving average MFCC music.
Keyword: segmental CDP, speech, music, segmentation, classification