• Seminars

   Brice GATINET  presents  In Girum:

This event will present In Girum, a compositional project in collaboration with ACIDS team I started during my residency at IRCAM. Pieces of this project devellop exclusively two Pure Data’s objects based on a machine learning's model called Variational Auto Encoders. This type of model is capable of learning a probability distribution of complex sound objects. During training, a list of input sounds is presented to the machine and it is asked to learn how to reconstruct them as precisely as possible. The quality of the model is estimated by a reconstruction error which evaluates the difference between the original and the generated sound. The layers between the input and the latent variables are considered as an encoder. In contrast, the layers between the latent variables and the output are called a decoder. The very low number of neurons in the central layer (named latent space) can be viewed as a bottleneck that forces the model to find a simplistic representation of the original data which is capable to reconstruct the original state of the data by itself. 

The name of this project ("Spinning around" in English) depicts this encoding/decoding operation and also evokes two main extra-musical references: 1) an evocation of the concept of circular time and  2) the verse constituted in palindrome ¨In girum imus nocte ecce and  consumimur igni "(We Spin Around the Night Consumed by the Fire) in reference in particular to the French activist Guy Debord.

Sound examples and enhanced specifications of the model are available here :

En poursuivant votre navigation sur ce site, vous acceptez l'utilisation de cookies pour nous permettre de mesurer l'audience, et pour vous permettre de partager du contenu via les boutons de partage de réseaux sociaux. En savoir plus.