Musical Instrument Recognition

The goal of this project is to separate three different instruments, namely guitar, cello and piano. The minimal goal is to train a Machine Learning algorithm with a moderate number of training samples in order to build a mathematical decision boundary to distinguish those three instruments. Certainly, the solution is supposed to be extensible in a way that without a lot of effort further instruments can be recognized. In the beginning the algorithm is trained by a tone with a frequency of 440Hz; in a musical environment called a’. The point of the separating approach, which is presented in this document, is to extract the ratio between the recorded frequency(440Hz) and all peaks of further resonating frequencies. The training is done by a classical SVM algorithm. A long-distance goal would be to develop a more robust system which can be relevant to various frequencies and to train the system with more instruments.