Identifikasi campuran nada pada suara piano menggunakan codebook

Abstract

It has been identified blend of tones, either as a single tone or a mix tones using the mel-frequency cepstrum coefficients (MFCC) as feature extraction and modeling of codebook for pattern recognition. The voices which used are the sound of piano and identified as 12 single-tones and 66 mix of tones recorded with 11 kHz on the duration of 1 second. The making of codebook is applied step by step, that is the codebook with the blend and the codebook of tones (single and mixed tones) which was developed using clustering techniques. The results of research showed that the optimum number of codewords is 20 with width of frames is 256 data, with an accuracy is 98.2%. However, there are some tones which are difficult to be identified, they are CC#, CD, CF, and A#B which has accuracy of each below 50%. For C# tone is more often recognized as C#D# tone, for the CC# tone is more often identified as C tone, for CD tone is more often recognized as C# tone, the CF tone is more often identified as C# tone and CF# tone, while for the A#B tone is more often identified with A# tone. Some errors happen in this recognition because those tones have similiar signal pattern so that they have the vector points which adjacent to each other.