Authors:

Emma Lombardo

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

Automatic music transcription (AMT) is a powerful tool that can help musicians in many ways. It can aid music education, creation, and production while also eliminating the time and resources needed to transcribe music by hand. But difficulties arise when considering all the pieces that make AMT models work. Datasets are often low resource, sounds and harmonies in audio overlap in frequency making separation difficult, musical attributes such as pitch and velocity are hard to infer from polyphonic music, and there are storage, network, and memory complexity constraints. The initial stage of research aimed to explore what AMT models exist through literature. We then tested different models to examine how they worked and observe their transcribed MIDI files. We also researched neural networks as they are common techniques for AMT. Utilizing the Multi- Task Multitrack Music Transcription (MT3) model, various simple and complex pieces were used to test the capabilities of the model. We found that the quality of the transcribed sheet music to be better in terms of note correctness and instrument identification when there were less instruments and when the piece was simple. Future work involves continuing our literature research and then taking and improving an existing model. Keywords: Music Transcription; Automatic Music Transcription; Polyphonic Music; Artificial Intelligence; Machine Learning

Source:

Purdue University / 2024

Topics:

No topics listed

Co-authors:

Emma Lombardo