Audio-to-Midi Similarity For Music Retrieval

Presenter: Fábio Goródscy

We are presenting strategies for processing audio for query-by-humming, having the goal of matching melodies from MIDI files to the melody hummed in audio files. An application of query-by-humming is defined as interface where an user can hum a melody as he or she remembers and the application brings melodies from a MIDI repertoire that has some degree of similarity, depending on what is expected by the user to be similar, so that the user can discover more information from the melody he hummed.

We are showing a brief revision of key concepts and examples of transcription using algorithms as Melodia and ASyMuT. We discuss metrics for matching these sound representations and show some results using the presented strategies.

When: June 4th, 2018

Where: Auditório Antonio Gilioli, IME/USP