Séminaire ETIS : Dogac Basaran

Titre du séminaire et orateur

The dominant melody estimation in Jazz music.

Dogac Basaran, IRCAM (UMR9912 - CNRS - Sorbonne Université).

Date et lieu

Mardi 12 mars 2019, 11h.

ENSEA, salle du conseil (salle 165).


Automatic dominant melody estimation (AME) from a polyphonic audio recording is a popular and rather challenging task in Music Information Retrieval (MIR). The difficulty is that the polyphonic accompaniment to the lead vocal/instrument follows the melody rhythmically and harmonically, in the sense that chord progressions will naturally contain the dominant F0 and/or its harmonics.

In this talk, we'll cover the current methods to solve this problem such as using handcrafted features and ad-hoc methods, machine learning methods such as non-negative matrix factorization (NMF) and deep neural networks (DNN) mostly using Convolutional and Recurrent networks. We'll focus on mostly DNN solutions and the analysis on the Jazz music, specifically on the improvisations or solo parts. Finally, we'll discuss about how to enhance such system for different tasks such as note transcription or pattern recognition.


Dogac Basaran received his Ph.D. degree from the Department of Electrical and Electronics Engineering at Boğaziçi University 2015. From November 2014 to October 2016, he worked as a research engineer at NETAS Telecommunications on multimedia communications. From October 2016 to January 2018, he was a postdoctoral researcher with the department of Image, Data and Signal Processing (IDS) at Télécom Paristech in Paris (France)  where he worked on speaker diarization and respresentation learning for audio. He is currently a postdoctoral researcher with IRCAM working on automatic melody extraction in the framework of the Dig That Lick project.

His research interests mainly include audio/music signal processing, Bayesian and approximate inference, machine learning for signal processing and deep learning methods.