The ARS modules, are one of the main results of the ANR project ARS (Analysis and tRansformation of Singing Style,http://ars.ircam.fr)

These modules integrate three deep neural networks : the RIU Separator, a deep neural network for singing voice extraction, FCN-F0 a fully convolution network for F0 estimation, and Vax, a singing voice alignment model based on the Adagio model supporting all Europeen and other languages. We will demonstrate the use of the ARS modules for analysis of singing voice in popular music.

ISiS 1.3.0 will become a major upgrade of Ircam’s Singing Synthesis software. Following our last presentation on ISiS 1.3 we have made significant progress towards the finalization of the song representation, a json file integrating all the information necessary for synthesizing a song. We will demonstrate a the ISiS viewer, a first prototype of a GUI editor for ISiS’s style parameters, and a first synthesis of the RK soprano voice, a new voice that has been developed in the Innovation Projet ISiS Voices and that has been made possible by the deep learning alignment model Adagio tat has been developed by Yann Teytaut, PhD student in the ANR project ARS.

Finally we will demonstrate the Circe plugin, an experimental investigation into converting the Circe model into a VST plugin for voice transformation.