Extraction of feature vectors for analysis of musical instruments

Music data analysis and retrieval has become a vital and challenging research field due to the increasing need to manage large amount of musical data present on multimedia. Music analysis deals with extracting various features, which describe the music signal, and use those features to process the music samples for various applications such as indexing, retrieval, automatic classification, etc. This paper presents analysis performed on Indian musical instrumental signal to extract different features in temporal, spectral, cepstral & wavelet domain. Analysis is performed on a large database consisting of around 700 sample files of five musical instruments specifically Indian musical instruments namely flute, dholaki, harmonium, santoor & sanai, belonging to woodwind, percussion, keyboard, string & brass category respectively. Feature vectors in the four domains are extracted for all files in the database. These feature vectors are further analyzed to observe their usefulness in discrimination amongst the instruments.