Дослідження транскрипції аудіо-даних в MIDI-формат з використанням нейронної мережі для розширення цільової аудиторії

Abstract

Магістерська кваліфікаційна робота виконана студентом групи КНМ-21 Ярошевським Дмитром Миколайовичем. Тема «Дослідження транскрипції аудіо-даних в MIDI-формат з використанням нейронної мережі для розширення цільової аудиторії». Робота направлена на здобуття ступеня магістра за спеціальністю 122 «Комп’ютерні науки». Метою магістерської кваліфікаційної роботи є розробка нейронної мережі для перетворення записаного звуку фортепіано в MIDI запис. Предметом дослідження виступають методи та засоби глибинного навчання для конвертації аудіозапису в MIDI-формат. Об’єктом дослідження є процес перетворення музичного запису в MIDI-формат. В результаті виконання дипломної роботи було розроблено нейронну модель нейронної мережі, яка перетворює звуковий запис в MIDI. Nowadays, more and more musical works appear on the Internet and, finally, in life in general. We notice how the mood and inspiration come when we listen to songs that are selected according to our preferences or situation. Many people who listen to many bright and wonderful works are so inspired that they decide to create their own work that will reflect their thoughts, feelings, events, depict the world around them. They sing a certain melody and then, if they see in it what is worth showing to others, they try to play on at least one instrument to remember or start the process of processing and improvement. Usually, the main melody of a work is recorded on keyboard instruments, such as a piano or grand piano. The piano allows you to transmit the best reproduced sound relative to the sound frequencies. This is due to the design features of the tool. The principle of playing the melody of the note is that when you press the key, the hammer strikes the string, which is tuned to play exactly the note and with the frequency you need, which leads to a melody with a fairly clear sound. Of course, the force of pressure also plays a role, because it conveys the strength of sound and appropriate intonation in the work. For example, it will take more time for a beginner to play such precise sounds on a violin or guitar, and it will not always be possible to convey such completeness of the given melody. As we know, any piano usually has 88 notes, which are located on 7 octaves. As you know, there are MIDI keyboards that have fewer keys just to record the main melody of the song. They are called synthesizers and are used by composers who have been working with sound for a long time. This kind of instrument is rare, so novice musicians or children studying in music schools are unable to obtain such an instrument, which means that there is a request to convert an audio recording recorded on a dictaphone into a MIDI recording for editing. MIDI recording contains information about the appearance of notes in relation to time, its duration, sound attack and many other data that help to reproduce sound recording on other devices, similar to the natural sound of the same composition on simple instruments.