Identification of Music Instruments from a Music Audio File

Abstract

The goal of this research is to design an intelligent system to recognize the specific musical instrument from an audio file and construct the notation text file correspondingly for the Music Instrument Digital Interface file. The scope was limited to the identification of two different instruments of the same pitch at a time and identification is done based on the analysis of the waveform pattern of instrument sound. Ten different musical instruments from a variety of musical instrument categories were selected to observe their waveform patterns. A sampling of different tones of sound patterns of the musical instruments was carried out. For this research, 12 tons of three different instruments in three different instrument categories were selected. From these three instruments, the same pitch sound combination of two instruments at a time was used to take the samples. There are two parts in the system; first is feature extraction of the sampled waveform and the second is feature classification. In the first part, feature extraction is done with mel frequency cepstral coefficient algorithm. Thirteen mel frequency cepstral coefficient features of the sampled waveform have been extracted and then the neural network has been trained using those coefficients. In the second part, classification is done with an artificial neural network. Its functionality was evaluated using confusion matrices. Based on the performance analysis of the result, the project was successful in detecting two different instruments in two different instrument categories played simultaneous time.