You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
Abstract<br />
The main goal of this thesis is the development and evaluation of an automatic<br />
musical instrument recognition system. Aim of the system is to recognize a musical<br />
instrument from a note’s signal, which signal is imported via a system’s microphone.<br />
The system is developed with techniques that are used in Speaker and Speech<br />
Recognition.<br />
The instruments, which are called to recognize are piano, xylophone, recorder,<br />
guitar and harmonica. The classifier that has been used is the Gaussian Mixture<br />
Models (GMM) and the features as for the training and testing data are MFCC, delta<br />
and delta-delta. Use of these features gives very good results and correct recognition<br />
of musical instruments (in percent) from the note’s signal, which imported to system<br />
via microphone, is 92.65%.Use of less features decrease the efficiency of the system.<br />
Also, changes in some parameters(such as framesize) decrease the efficiency. The<br />
parameters that were changed as well as the results of correct recognition of the<br />
system are reported in Chapter 6.<br />
Keywords:<br />
Musical instruments, Musical instruments Recognition, Digital Signal Processing,<br />
Gaussian Mixture Models, MFCC, delta, delta-delta, feature extraction, xylophone,<br />
recorder, guitar, piano, harmonica, pattern recognition, ear physiology.