Protecting Copyright Ownership via Identification of Remastered Music in Radio Broadcasts

Abstract

Music identification in radio broadcasts have different use cases like playlist generation and protecting copyright ownership of artistes. In real world environments, music is often remastered by radio channels to fit in to limited air times. Mostly those remastering includes time stretching, pitch shifting and etc. Therefore robustness of the music identification method plays a crucial role. In this dissertation, we propose to use Scale Invariant Feature Transform (SIFT) on Short Time Fourier Transformation (STFT) spectrogram to extract audio descriptors. Experiments show that SIFT descriptors exhibit robustness against audio distortions such as time stretching and pitch shifting. Finally a ratio based threshold is used to differentiate identified and non-identified states.