Training a filter for noise reduction in a hearing device
US10897675B1 · kind B1 · utility
Assignee
Inventor
Key dates
| Filing date | Aug 14, 2019 |
| Grant date | Jan 19, 2021 |
| Priority date | — |
| Expiry date | Aug 14, 2039 |
Classification
- Technology area (CPC H)Electricity
- CPC primaryH04R2225/67
- WIPO fieldComputer technology
- WIPO sectorElectrical engineering
Abstract
The disclosed technology relates to training a neural network and using that training to design a filter for the hearing device. The training can include training a generative adversarial network (GAN) to determine whether a signal includes intelligible speech, wherein the GAN includes a generator and discriminator. The generator can provide a mixed noise and speech signal to the discriminator can output a signal that reduces noise in the mixed signal. The technology can then determine whether the output signal is intelligible based on a speech metric function or based on feedback from a hearing care expert. The generator and the discriminator can continue to train based on this feedback. And then the trained network can use the settings of the discriminator or information from the design of the discriminator to generate a noise filter for a hearing device.
Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.