Sonova Ag

Sonova

Samuel Bucher

Hans-Ueli Roeck

Fethi Cherigui

Xavier Gigandet

Jean-Louis Durrieu

Hannes Wuethrich

Markus Hofbauer

A processing chip for processing audio signals using at least one deep neural network (DNN) in a hearing device comprises a first compute unit having a hardware architecture adapted for processing one or more convolutional neural network layers of the at least one DNN, a second compute unit having hardware architecture adapted for processing one or more recurring neural network layers of the at least DNN, a control unit for directing the first and second compute units when to compute a respective layer of the at least one DNN, a shared memory unit of storing data to be processed in respective layers of the at least one DNN, and a data bus system for providing access to the shared memory unit for each of the first and the second compute unit.

Processing chip for processing audio signals using at least one deep neural network in a hearing device

This invention relates to a method of managing adaptive feedback cancellation in a hearing device comprising at least a microphone (); a receiver (); and a signal processing circuitry, configured to receive from said microphone () an input signal (X, X, X) and to provide said receiver () with an output signal (Y, U, U). An external acoustic feedback path () defined by feedback sound (fb) traveling from the receiver () to the microphone () is represented by an external feedback path transfer function (). The signal processing unit comprises at least a gain unit (); a feedback canceller unit () comprising an adaptive filter element () configured to adaptively accommodate changes in said external acoustic feedback path transfer function (); and a frequency shift unit () configured to stabilize an adaptation of the feedback canceler unit (). The method according to the present invention comprises the steps of: estimating the external acoustic feedback path transfer function () by modeling/deriving a first estimated feedback path transfer function () to reflect the external acoustic feedback path transfer function (), by said feedback canceler unit (); compensating the input signal of the hearing device (X), based on the first estimated feedback path transfer function (), thereby generating a compensated input signal (X); providing at least part of the signal processing unit with the compensated input signal (X). The method further comprises the steps of: generating a probe signal (W), by an adaptation control block (); injecting the probe signal (W) into the output signal (Y) of the hearing device and letting the probe signal (W) be fed back to the microphone () through the external acoustic feedback path (); modeling/deriving at least a reference estimated feedback path transfer function (), based on a relation between the input signal (X) and the probe signal (w); comparing said at least a reference estimated feedback path transfer function () with said first estimated feedback path transfer function (), by a comparison unit () of said adaptation control block (); and controlling the adaptive filter element () and the frequency shift unit () based on a comparison between such at least a reference estimated feedback path transfer function () and the first estimated feedback path transfer function ().

Method of managing adaptive feedback cancellation in hearing devices and hearing devices configured to carry out such method

Growing community of inventors

Rapperswil, Switzerland

Samuel Bucher