Effects of Beam Search Algorithm on Machine Generated Music

Sammanfattning: In this thesis, we are trying to understand the correlation between the beam-search algorithm and plagiarism. We are training a Long Shortterm Memory (LSTM) network, on a larger data-set of music transcriptions. During sampling from the model, we are employing the beam-search algorithm to generate a specific kind of traditional Irish dance music i.e. double-jig. Our motivation to generate Irish dance music comes from The AI Music Generation Challenge 2020 [13]. After generating music transcriptions from the model we are employing a text-based plagiarism detection method. Plagiarism detection in the machine-generated samples is performed against the training data-set. We believe results from plagiarism detection on machine-generated music transcriptions will help us to understand how the beam-search algorithm affects the plagiarism in the machine-generated music transcriptions.