| Tartalmi leírás: |
In this thesis I apply a convolutional neural network to classify a large set of publicly available audio files into classes depending on the instrument played in the file. In order to do this, a logarithmic Fourier transform is applied onto smaller segments of the audio files, and a 2-D map of the frequencies is created which is passed to the neural network. Using this process, it is possible to achieve an accuracy of 90.7% with 4 classes, or an accuracy of 75.7% with as many as 8 classes.
|
