Instrument Classi cation for Cochlear Implant Filtering using Convolutional Neural Networks

Abstract

In this thesis, Convolutional Neural Networks are used to classify instru- ments in audio les containing musical instruments. In particular, music enjoyment for Cochlear Implant users is central in this thesis. Therefore, there are di erent frequency lters applied in the preprocessing, to further explore the optimal preprocessing when frequency space is limited. The results show that, in line with previous research, there is a certain range that is particularly important in instrument recognition. However, in this research this range di ers, possibly because this research is focused on music whereas the original research is based on speech. The classi cation when ltered on this range outperforms the classi cation when presented with the full spectrum.