Academic project
Composer-Specific Analysis of Automatic Piano-Roll Music Transcription
Study of composer-specific deep learning models for transcribing music audio into piano-roll representations.
The project investigates automatic music transcription: converting recorded audio into a piano-roll representation of musical notes. Its hypothesis, as stated in the Lattes record, is that deep learning models trained separately for each composer may perform differently from a single model trained across multiple composers.
The proposed method uses audio-derived spectrograms as inputs to neural networks. Planned experiments employ MusicNet and assess precision, recall, and accuracy. The source describes a research plan and does not, by itself, establish experimental outcomes.
Team listed in Lattes: Jefferson Oliveira Andrade, Karin Satie Komati (project lead), Leonardo Veronez Simões, Eduardo Henrique Próspero Souza. Lattes lists this project as “ongoing”; its current status has not been independently checked. Support listed in Lattes: Fundação de Amparo à Pesquisa do Espírito Santo (scholarship).