Honors College Thesis
 

Transcribing Solo Piano Performances from Audio to MIDI Using a Neural Network

Public Deposited

Downloadable Content

Download PDF
https://ir.library.oregonstate.edu/concern/honors_college_theses/ff365d34n

Descriptions

Attribute NameValues
Creator
Abstract
  • Automatic Music Transcription is a growing area of interest in Music Information Retrieval, and recent research has shown promise using onset detection on spectrograms. We propose a deep learning model that takes raw audio as input in order to transcribe a solo piano performance from audio to MIDI without complicated audio pre-processing and translation to spectrograms. This project focuses on the representation of piano audio as MIDI, but the contributions of this work will have applications in the area of piano sheet music creation. In this work we show that a bidirectional long short-term memory neural network can be used to transcribe a solo piano performance with high levels of both pitch and onset accuracy. Despite our unique approach, our model outperforms early models in this field and approaches competitiveness with top-performing models of the last two years. Our model is the only model in recent years to use raw audio as opposed to spectrograms or constant Q transforms. Because we use raw audio, we are able to capture resonance that exists within an audio recording that is not present in MIDI recordings. We also correctly model the behavior of the sustain pedal, despite having difficulty transcribing it correctly. These results show the promise of simpler models than are currently used for Automatic Music Transcription and bode well for the use of raw audio in music transcription.
Resource Type
Date Issued
Degree Level
Degree Name
Degree Field
Degree Grantor
Commencement Year
Advisor
Committee Member
Non-Academic Affiliation
Rights Statement
Publisher
Peer Reviewed
Language

Relationships

Parents:

This work has no parents.

In Collection:

Items

Thumbnail Title Date Uploaded Visibility Actions
file details EbertVictoria2022.pdf 2022-05-30 Public Download