Automatic Note Recognition and Generation of MDL and MML using FFT

Increase in the number of online released music tracks means there is a need to improve the performance of existing Music Information Retrieval (MIR) systems, e.g., music search engine, and to develop a music version of Turn-it-in tool, a plagiarism checking system that evaluates the similarity between two pieces of music tracks. In our previous papers, we have proposed a symbolic coding scheme, named Music Definition Language (MDL) and Music Manipulation Language (MML), to describe music pieces; and designed corresponding algorithms to compute the similarity scores. However, the potential of MML/MDL won't be released if existing music pieces have to be manually transformed to their MML/MDL representations. This paper is focused on the automatic transformation of audio tracks, either analogue or digital signal, into MDL/MML representations. Following the case study approach, we first use MatLab to generate an MP3 file from a predefined MDL and MML representation, which correspond to the Net-based online music. Next, we adopt Fast Fourier Transform (FFT) to convert the MP3 file back to the MDL and MML file and check how accurate we can obtain via the similarity score. To achieve this purpose, the following features need to be extracted: frequencies, amplitude, time the note has been played and the duration. The experiment shows 93.63% contour melody similarity score and 100% for the rhythm have been achieved compared to the original file.