Hearing device comprising a detector and a trained neural network

A method of selecting optimized parameters for a neural network for use in a portable hearing device, the method comprising:

• providing a portable hearing device to be used by a particular user, the hearing device comprising a neural network processor adapted to implement a neural network, the neural network being configured to receive an input vector and to provide an output vector as a specific non-linear function of the input vector;
• mounting the hearing device at or on the user;
• communicatively connecting said hearing device to an auxiliary device, and entering an adaptation-mode of operation in response to said connection, said auxiliary device being provided with multiple pre-trained neural networks, where each pre-trained network is considered a candidate for the user, and wherein each pre-trained neural network has been trained on fully or partially different training data;
• and
• while operating according to said adaptation-mode,
• providing, via the hearing device, at least one electric input signal representing sound in the environment of the user wearing the hearing device;
• processing said at least one electric input signal and providing a multitude of feature vectors, each being representative of a time segment of said at least one electric input signal, the auxiliary device using the feature vectors to select from said multiple pre-trained neural networks one tailored to the user's needs;
• and
• receiving, by the hearing device, data from the auxiliary device to implement the selected one of the multitude of pre-trained neural networks as the neural network in the neural network processor.

A hearing device configured to be located at or in an ear, or to be fully or partially implanted in the head, of a user, the hearing device comprising:

• an input transducer comprising at least one microphone for providing at least one electric input signal representative of sound in the environment of the hearing device;
• a processor for processing said at least one electric input signal and providing a multitude of feature vectors, each being representative of a time segment of said at least one electric input signal, wherein
• the processor comprises a neural network processor adapted to implement a neural network for implementing a detector, or a part thereof, configured to provide an output indicative of a characteristic property of the at least one electric input signal, the neural network being configured to receive said multitude of feature vectors as input vectors and to provide corresponding output vectors representative of said output of said detector, or a part thereof, in dependence of said input vectors,
• the processor is configured to receive said output vectors, and the processor is configured to process said output vectors and to provide a resulting signal;
• a transceiver comprising a transmitter and a receiver for establishing a communication link to another part or device or server, said communication link allowing transmission and reception of data to and from, respectively, said another part or device or server, at least in a particular adaptation-mode of operation;
• and
• a selector for, in said particular adaptation-mode of operation, routing said feature vectors to said transmitter for transmission to said another part or device or server and, in a normal mode of operation, to route said feature vectors to said neural network processor for use as inputs to said neural network,
• wherein the hearing device is configured to, in said particular adaptation-mode of operation, receive optimized node parameters from said another part or device or server, and to apply said optimized node parameters to said nodes of the neural network to thereby implement an optimized neural network in the neural network processor,
• wherein the optimized node parameters have been selected among a multitude of sets of node parameters for respective candidate neural networks according to a predefined criterion in dependence of said feature vectors, and
• wherein the detector or part thereof implemented by the neural network is a wake-word detector, a keyword detector, a preferred speaker detector, an on/off detector, or an own-voice detector.