About
AliceCohen-Hadria
Machine learning and signal processing PhD student
About
I am interested in using deep learning method to estimate any musical description. Particularly, I’ve worked on musical structure estimation and voice processing with Convolutional Neural Network (CNN). My work fits in Music Information Retrieval (MIR) field, from signal processing to any neural network methods. I am currently employed by IRIT (Toulouse Institute of Computer Science Research) for a postdoc on pronunciation errors detection.
CV
Curriculum Vitae
CV
1 - Work Experiences
- Postdoc @IRIT - Automatic Speech Recognition for Computer-Assisted Language Learning. Supervised by Isabelle Ferrané, Thomas Pellegrini and Julien Pinquier (IRIT, France), Lionel Fontan and Maxime Le Coz (Archean LABS, Montauban, France).
- PhD @IRCAm - Music Description Estimation with Deep Learning. Supervised by Geoffroy Peeters. Defense video Defense slides
- Visiting Scholar @NYU - Voice anonymization in urban sounds. Supervised by Mark Cartwright, Brian McFee and Juan Bello.
- Internship @LIP6 - Generic programming for combinatorial classes in Ocaml. Supervised by Matthieu Dien.
3 -Meeting Organization
- Co-chair of Unconference at ISMIR 2018, Paris, France. 400 participants, chairing and organization of debates on MIR subjects.
- ISMIR 2018 Local committee organization, Paris, France. Meeting of the International Society of Music Information Retrieval, 450 participants.
2 - Teaching
4 - Education
- Master (2nd year) in Computer Science and Signal Processing applied to Music @IRCAM Link
- Master (1st year) in Computer Science - Distributed agents, Robotics, Operational research, Interaction and Decision. UPMC 2011- 2014
- Bachelor degree in Computer Science, With Honors, UPMC
Papers
Academic publications
Papers
Voice anonymization in urban sound recordings
In this paper, we propose an anomization method that obfsucate the content and msask the speaker identity, while preserving the acoustic scene.
Source separation & Data augmentation
In this paper, we present our data augmentation algorithm, tailored for singing voice extraction
Structure estimation
We investigated different input representation to detect music structure boundaries.
DALI dataset - First paper
We constructed the largest dataset of audio and lyrics aligned, using singing voice detection with deep learning.
DALI dataset - Second paper
Second paper on the DALI dataset, presenting 2nd version. An human evaluation of the produced alignment is also presented.