Mathias Quillot

Background

About Mathias Quillot

I am passionate about computer science and more particularly Artificial Intelligence (machine learning). During my PhD at the computer science laboratory of Avignon (LIA), I had the opportunity to work on the problem of recommending voice actors by deep learning for the movie industry. Now, I am working as Senior Applied Scientist at Oracle. My PhD years allowed me to develop a large set of skills. I mainly worked with Kaldi or PyTorch to train and infer speaker recognition models. I also used PyTorch and TensorFlow to train voice similarity systems and Pyannote to exploit a pre-trained model of diarization. In order to train my systems, I used the LIA's cluster of computers where slurm scheduler is installed. My research has mainly revolved around the question of voice representation. The objective of my work was to highlight the character (role) aspects of the voice in a particularly complicated context where it was difficult to differentiate the speaker from the role. To achieve these objectives, I proposed a protocole that assess the existence of character aspects independent of the speaker in a set of representation. Lastly, I also enjoy contributing to the life of the institution in which I work. When I was at the LIA, I organized reading challenges, bimonthly meetings to present scientific papers or works on speech processing (which we called language meeting). I was also an elected doctoral student at the scientific council of my laboratory and university and participated to economic and strategic meetings.