Method of operating an audio device system and audio device system

An internet server comprising a downloadable application that may be executed by a computing device, wherein the downloadable application is adapted to cause at least one of the following method steps to be carried out:

• passing a frame of an audio signal xt through a neural network encoder and hereby obtaining a latent encoding zt,
• manipulating said latent encoding zt to provide a transformed latent encoding zt′ by using a model configured to transform said latent encoding zt, wherein said step of manipulating said latent encoding Zi to provide a transformed latent encoding zt′ is adapted to provide at least one of:
• noise suppression;
• change of at least one speech characteristic selected from a group comprising:
• spectral speech characteristics, temporal speech characteristics, enunciation, pronunciation, pitch and phoneme emphasis;
• and
• at least one of sound source separation, removal, suppression and enhancement;
• using a forecasting model to provide a prediction of a future transformed latent encoding zt+k′ based at least on said transformed latent encoding zt′;
• passing said predicted future transformed latent encoding zt+k′ through a neural network decoder and hereby providing a predicted future transformed electrical output signal {tilde over (s)}t+k.

A non-transitory computer readable medium carrying instructions which, when executed by a computing device causes at least one of the following method steps to be carried out:

• passing a frame of an audio signal xt through a neural network encoder and hereby obtaining a latent encoding zt,
• manipulating said latent encoding zt to provide a transformed latent encoding zt′ by using a model configured to transform said latent encoding zt, wherein said step of manipulating said latent encoding zt to provide a transformed latent encoding zt′ is adapted to provide at least one of:
• noise suppression;
• change of at least one speech characteristic selected from a group comprising:
• spectral speech characteristics, temporal speech characteristics, enunciation, pronunciation, pitch and phoneme emphasis;
• and
• at least one of sound source separation, removal, suppression and enhancement;
• using a forecasting model to provide a prediction of a future transformed latent encoding zt+k′ based at least on said transformed latent encoding zt′;
• passing said predicted future transformed latent encoding zt+k′ through a neural network decoder and hereby providing a predicted future transformed electrical output signal {tilde over (s)}t+k.