Modern music production relies on digital devices and software. However, music production still involves process of playing instruments, which requires instrument players to accurately play the designated notes. Such process requires repetitive practise and recording. Therefore, researchers have been looking for different ways of music information retrieval to analyse music in audio form algorithmically. Converting audio to MIDI data is one of the important research in the music information retrieval field, which can increase the efficiency of music production as well as analytical tools for musicians.

The objective of this thesis is to investigates algorithms to convert audio directly to MIDI in real-time, particularly for human voice or potential instrument sound. Different t...[ Read more ]

Modern music production relies on digital devices and software. However, music production still involves process of playing instruments, which requires instrument players to accurately play the designated notes. Such process requires repetitive practise and recording. Therefore, researchers have been looking for different ways of music information retrieval to analyse music in audio form algorithmically. Converting audio to MIDI data is one of the important research in the music information retrieval field, which can increase the efficiency of music production as well as analytical tools for musicians.

The objective of this thesis is to investigates algorithms to convert audio directly to MIDI in real-time, particularly for human voice or potential instrument sound. Different techniques which are crucial to audio-to-MIDI conversion are discussed and compared, including pitch detection, Cepstrum analysis, note segmentation algorithms and musical pitch model.