Training a filter for noise reduction in a hearing device

A computer program product including instructions that when executed by a processor or processors, cause a device or devices to perform operations, the operations comprising:

• training a generative adversarial network (GAN) to determine whether a signal includes intelligible speech,
• wherein the GAN includes a generator and discriminator,
• wherein training includes:
• mixing, via the generator, a speech sample from a speech sample database with a noise sample from a noise sample database to produce a first mixed signal,
• providing the first mixed signal to the discriminator;
• generating, via the discriminator, a first output signal that is based on the discriminator reducing noise in the first mixed signal;
• determining whether the first output signal has speech intelligibility based on a speech metric function and/or based on feedback from a hearing care expert;
• providing information regarding the speech intelligibility of the first output signal to the generator and the discriminator;
• based on the provided information regarding the speech intelligibility of the first output signal, adjusting mixing settings of the generator and providing a second mixed signal to the discriminator based on the adjusted mixed settings;
• generating, via the discriminator, a second output signal that is based on the discriminator reducing noise in the second mixed signal;
• and
• based on the trained GAN and filter values of the discriminator, providing a filter to a hearing device,
• wherein the filter is configured to reduce noise in a hearing device signal.

A system for providing a filter to a hearing device, the system comprising:

• a generative neural network;
• a generator configured to produce a mixed signal based on mixing a speech sample with noise sample;
• a discriminator configured to produce an output signal based on discriminating between a speech portion of a signal and a noise portion of a signal;
• and
• a speech metric unit configured determine whether the output signal includes intelligible speech;
• and
• a filter generator configured to communicate with the discriminator and configured to generate filter settings for a filter for a hearing device,
• wherein the filter settings are associated with settings that the discriminator uses to discriminate between the speech portion of the signal and the noise portion of the signal.