Verfahren zum Trainieren eines Neuronalen Netzes

Details

Veröffentlichungs-Nr.

EP4814955

Anmeldetag

24. März 2026

Veröffentlichung

30. September 2026

Rechtsraum

EP

Abstract

A method for training a neural network is disclosed. The method comprises receiving, by a sound database, a training input. The training input comprises a speech component and a noise component. The method comprises receiving, by the sound database, a training target comprising the speech component. The method comprises providing, by a high-resolution analysis filter bank, a high-resolution training sub-band input based on the training input. The method comprises determining, by the neural network comprising a plurality of neural network parameters, one or more sub-band filters based on the high-resolution training sub-band input. The method comprises providing, by a low-resolution analysis filter bank, a low-resolution training sub-band input based on the training input. The method comprises providing, by filtering the low-resolution training sub-band input using the one or more sub-band filters, a training sub-band output. The method comprises providing, by the low-resolution synthesis filter bank, a training output. The method comprises training the neural network by updating, by a network optimizer, the neural network parameters based on reducing a neural network loss between the training output and the training target.

Anmelder

Firma

Oticon A/S

Land

🇩🇰 Dänemark