Optimised Quantisation Of Latent Space In Neural Audio Encoding

Abstract: The invention relates to a method for encoding an audio signal comprising an analysis (E401) of the audio signal by an analysis neural network in order to obtain a latent representation of the audio signal and in which the encoding of the latent representation is carried out according to the following steps: applying (E404) an orthogonal linear transformation (T) to a latent vector (z) representative of the latent representation of the audio signal or to a preprocessed latent vector (z') in order to obtain a transformed latent vector (y); quantising (E405) the transformed latent vector (y) obtained thereby. The invention also relates to a corresponding method for decoding an audio signal and to a corresponding encoding device and to a corresponding decoding device.

Patent Information

Applicants

Inventors

Specification

Documents

Application Documents