Music is social or cultural activity, an art form whose medium is soundand silence.
Rhythm (and its associated concepts tempo, Pitch (which governsmelody and harmony), meter, and articulation), the sonic qualities of timbreand texture (which are sometimes termed the “color” of a musicalsound) and dynamics (loudness and softness) are the common elements of music. Witha vast range of instruments, music is performed. Various instruments liketabla, harmonium, flute or violin are used in north Indian classical music. Toidentify type of musical instruments and singer voice, in the recent past,varies researches have been carried out.
Musical instrument identification/classificationis a subdomain of music information retrieval (MIR) and analysis. There arevarious ways to classify the musical instruments. In this project our aim is toidentify specific musical instrument based on the classification of musicalinstruments from the sound generated through them. For example ten flutes areplayed, and then we should identify the specific number of flute among all theflutes played. We are considering Monophonic music for this research.
In this,the musical instrument’s audio sample is recognized for its basic type ofmusical instrument. The audio input is preprocessed with respect to noise andthen normalized by amplitude. The audio feature extraction unit extracts theaudio attributes of the musical instrument in terms of audio descriptors. Theseaudio descriptors together give entire data space. Feature extractors such asMel Frequency Cepstrum coefficient (MFCC) as one feature space and combinationof MFCC with other Timbral audio descriptors as another feature space areanalyzed together. The Musicalinstrument classifier has ‘Training’ and ‘Testing’ phases that makes use ofVector Quantization to generate codebook. This codebook contains feature sampleper musical instrument class which will be used during the testing phase.
K-means being non-hierarchical method initially takes the number of componentsof the population equal to the final required number of clusters. We attempt touse K means classifier based on the technique of vector Quantization foridentification of a musical instrument and K nearest neighbor (KNN) as one thestatistical classifiers to identify the musical instrument. The finalconclusion will be based on the performance evaluation of both of theseclassifiers.