Hearing device comprising a detector and a trained neural network

Abstract

The present application discloses a hearing device comprising a detector and a trained neural network, wherein the hearing device comprises: an input converter; a preprocessor for processing at least one electrical input signal and providing a plurality of feature vectors; a neural network processor, which is suitable for implementing a neural network for implementing the detector and is configured to provide an output indicating a characteristic characteristic of at least one electrical input signal; a postprocessor, configured to receive and process the output vector and provide a resulting signal; a transceiver; a selector, which is used to send the feature vector to a transmitter in an adaptive operating mode and then to another part or another device or server, and to send the feature vector to the neural network processor in a normal operating mode; the hearing device is configured to receive optimized node parameters in the adaptive operating mode and apply them to nodes of the neural network, wherein the optimized node parameters have been selected from a plurality of sets of node parameters according to a predetermined criterion based on the feature vector.

Description

Technical Field

[0001] The present application relates to a hearing device such as a hearing aid comprising a detector for detecting a certain acoustic environment, such as a voice detector for detecting specific keywords for a voice control interface. The present application also relates to a scheme for personalization of hearing device parameters. Background Art

[0002] Detectors such as environment detectors, own voice detectors or keyword detectors usually require parameter adjustment. Detectors such as detectors providing a decision or providing a probability of one or more estimated parameter values ​​or aforesaid estimated parameter values ​​can for example be implemented using or based on supervised learning, for example using a neural network architecture in full or in part. Since the architecture of a neural network is very general, a neural network usually requires many parameters such as weights and bias parameters. Due to the limited memory in the hearing device, it is desirable to implement a neural network with as few parameters as possible. Furthermore, due to limited bandwidth during programming, it is desirable to transmit as few parameters as possible to the hearing instrument. Summary of the invention

[0003] Hearing Devices

[0004] In a first aspect of the present application, a hearing device is provided, which is configured to be located at or in an ear of a user or to be fully or partially implanted in a user's head. The hearing device comprises:

[0005] - an input transducer comprising at least one microphone for providing at least one electrical input signal representative of sounds in the environment of the hearing device;

[0006] - a preprocessor for processing the at least one electrical input signal and providing a feature vector (e.g. a plurality of feature vectors) representing a time period of the at least one electrical input signal;

[0007] - A neural network processor, which is suitable for implementing a neural network for implementing the detector or a part thereof, for example configured to provide an output indicative of a characteristic characteristic of at least one electrical input signal, the neural network comprising an input layer, an output layer and a plurality of hidden layers, each layer comprising a plurality of nodes, each node being defined by a plurality of node parameters, the neural network being configured to receive the feature vector (or the plurality of feature vectors) as an input vector and to provide an output vector (or a corresponding output vector) representing the output of the detector or a part thereof based on the input vector.

[0008] The hearing device may further comprise a signal post-processor configured to receive the output vector and at least one input signal representing a sound in the environment, and wherein the signal post-processor is configured to process the at least one input signal representing a sound according to the output vector and provide a processed output, i.e. a resultant signal. The hearing device may further comprise a transceiver comprising a transmitter and a receiver for establishing a communication link to another part or another device or a server, the communication link enabling data transmission to and reception from another part or another device or a server at least in an adaptive operating mode. The hearing device may further comprise a selector for sending the feature vector to the transmitter in the adaptive operating mode for transmission to another part or another device or a server, and for sending the feature vector to the neural network processor in the normal operating mode for use as an input to the neural network. The hearing device is configured to receive optimized node parameters from another part or another device or a server in the adaptive operating mode and apply the optimized node parameters to the nodes of the neural network so as to implement the optimized neural network in the neural network processor, wherein the optimized node parameters have been determined according to the feature vector (e.g. selected from a plurality of sets of node parameters for corresponding candidate neural networks according to a predetermined criterion).

[0009] The “other part” may be, for example, another part of the hearing device (e.g., integrated with the hearing device, or physically separated from the hearing device but communicating with the hearing device). The “other device” may be, for example, a separate (auxiliary) device, such as a separate wearable device such as a remote control device or a smart phone, etc. The “server” may be, for example, a computer or storage medium that is remote from the hearing device, for example, accessible via a network such as the Internet.

[0010] In a second aspect, the hearing device itself may include a plurality of pre-trained candidate neural networks, such that the adaptive mode can only be performed in the hearing device, e.g. the adaptive mode is run during daily use of the hearing device. The hearing device may include a plurality of pre-trained candidate neural networks (from which an appropriate neural network evaluated as performing best for the user in question is selected for use in the hearing device), such that the adaptive mode can only be performed in the hearing device (e.g. see Figure 8 ). The hearing device may for example be configured to operate in an adaptive mode during everyday use of the hearing device.

[0011] Thereby an improved hearing device may be provided.

[0012] The detector may be configured, for example, to identify a characteristic characteristic of the at least one electrical input signal. The "characteristic characteristic" may, for example, relate to a characteristic relating to the nature of the sound currently picked up by the at least one microphone and represented by the at least one electrical input signal. The "characteristic characteristic" may, for example, relate to the acoustic properties of the current acoustic environment. The "characteristic characteristic" may, for example, relate to speech components and noise components, self-speech and other speech, single-speaker environments and multi-speaker environments, speech and music, the content of a given speech sequence, etc.

[0013] The predetermined criterion may be related to minimizing a cost function with respect to the output vector. When the plurality of feature vectors are extracted from a time period of the at least one electrical input signal having known properties, the predetermined criterion may be based, for example, on the performance of the neural network in terms of positive, false positive, true rejection and false rejection of the output vector.

[0014] Multiple sets of node parameters for corresponding candidate neural networks can be optimized for different categories of people exhibiting different acoustic properties.

[0015] The neural network processor may be particularly suitable for performing calculations of a neural network. The neural network processor may form part of a digital signal processor. The digital signal processor may include a neural network kernel that is particularly suitable for performing operations of a neural network. The node parameters of a particular node may include, for example, weight parameters and (w) bias parameters (b). The nonlinear function (f) associated with each node may be exactly the same for all nodes, or may be different between layers or between nodes. The nonlinear function (f) may be represented, for example, by a sigmoid function, a rectified linear unit (ReLU) or a softmax function. Different layers may have different nonlinear functions. However, the parameters of the nonlinear function may also be learned (e.g., included in an optimization process). This is the case for the rectified linear unit of the parameters. In other words, the parameters of the nonlinear function (f) may form part of the node parameters.

[0016] A given neural network can be defined by a given set of node parameters.

[0017] The detector (or part thereof) implemented by the neural network may for example include a wake-up word detector, a keyword detector or a preferred speaker detector (spouse detector). Alternatively, the detector (or part thereof) implemented by the neural network may for example also include a correlation detector, a level estimator, a modulation detector, a feedback detector, a voice detector such as an own voice detector, a speech intelligibility detector of the current electrical input signal or a signal derived therefrom. The output of the detector may include an estimate of the value of a specific parameter or property or content of the electrical input signal or a probability of such an estimated value.

[0018] The detector may comprise a neural network processor or form part of a neural network processor. The detector may comprise a pre-processor. The detector may comprise a post-processor.

[0019] The term "sound in the environment" may refer to any sound in the environment surrounding the user wearing the hearing device, which can reach the at least one microphone and can be detected as an electrical input signal. "Sound in the environment" may include, for example, any sound produced by a person in the environment and any sound produced by a machine or device. "Sound in the environment" may include a specific word or multiple specific words of a specific person (such as the user). "Sound in the environment" may include natural sounds or background noise.

[0020] The hearing device may comprise a neural network controller connected to the neural network processor for receiving the optimized node parameters in an adaptive operating mode and applying the optimized node parameters to the nodes of the neural network so as to implement the optimized neural network in the neural network processor. The hearing device may comprise candidate neural networks. The candidate neural networks may be of the same type, such as a feed-forward, recursive or convolutional neural network. The candidate neural networks may comprise different types of neural networks. The candidate neural networks may differ in the number of layers and nodes, such as some candidate neural networks having more layers (and / or more nodes in a given layer) than other candidate neural networks. Thereby the complexity of the neural network may be adapted to a specific user.

[0021] The post-processor may be configured to provide a "decision" based on the output vector of the neural network. The post-processor may be configured to determine a resulting (estimated) detector value based on the output vector of the neural network, for example based on the probability of a plurality of different (estimated) detector values. The decision or resulting estimated detector value may for example be or include a specific wake-up word or command word of a voice control interface or a specific value of a voice detector such as an own voice detector, for example indicating the presence or absence of voice, such as the user's own voice.

[0022] The output of the decision unit (post-processor), i.e. the resulting signal, may for example be a command word or sentence or a wake-up word or sentence for activating a voice control interface. Alternatively or additionally, the output of the decision unit (post-processor) may be fed to a transmitter (or transceiver) to be transmitted to another device and further processed there, e.g. to activate a personal assistant (e.g. of a smartphone etc.). The transceiver may for example receive a response from another device, e.g. from a personal assistant. The response may for example be used to control the hearing device, or it may be played to the user via an output transducer (SPK) of the hearing device.

[0023] A hearing device may include a sensor for sensing a characteristic of a user or an environment of the hearing device and providing a sensor signal representing a current value of the characteristic of the environment. In the present specification, an environmental characteristic of a hearing device may include a characteristic (such as a parameter or state) of a user wearing the hearing device. In the present specification, an environmental characteristic of a hearing device may include a characteristic of a physical environment such as an acoustic environment of the hearing device. In the present specification, the term "sensor" may refer to a device that provides an output signal reflecting the value of a current physical parameter that the sensor is suitable for measuring. For example, a temperature sensor gives as an output the current temperature of the sensor environment. A hearing device may include a plurality of sensors. A sensor may consist of or include the following sensors: a motion sensor (such as an accelerometer), a magnetometer, an EEG sensor, an EOG sensor, a heart rate detector or a temperature sensor, etc. A voice activity detector (VAD) or an OV detector may also be considered a sensor.

[0024] The hearing device may be configured such that the sensor signal is an input to the pre-processor.

[0025] The preprocessor may be configured to process at least one electrical input signal and the sensor signal to provide a feature vector. The sensor signal may be used to qualify (e.g. exclude certain time periods (or frequency bands) of) the at least one electrical input signal before generating the feature vector. Alternatively or additionally, the sensor signal (or features thereof) may be included in the feature vector provided by the preprocessor as an input vector to the neural network.

[0026] The hearing device may comprise an output unit for presenting the processed output signal to a user as a stimulus perceivable as sound. The output unit may comprise an output transducer. The output unit (e.g., output transducer) may comprise a loudspeaker of an air conduction hearing device, a vibrator of a bone conduction hearing device, or a multi-electrode array of a cochlear implant hearing device. The output unit may comprise a transmitter for transmitting the resulting (detector) signal and / or the output (audio) signal to another device.

[0027] The hearing device may comprise an analysis filter bank for converting a time domain input signal into a plurality of sub-band signals thereby providing an input signal in a time-frequency representation (k, l), wherein k and l are frequency and time index, respectively. For each electrical input signal and / or sensor signal, the input transformer may comprise an analysis filter bank. The hearing device may comprise an analysis filter bank for converting a time domain sensor signal into a plurality of sub-band signals thereby providing a sensor signal in a time-frequency representation (k, l), wherein k and l are frequency and time index, respectively. The eigenvector may be provided in the time-frequency representation. If reconstruction of the time domain signal is not required (not required if only the eigenvector is used for the detector), the filter bank may be downsampled by a factor higher than the critical downsampling. We may also use channels of a smaller subset of the available channels of the filter bank, and we may add the channels together. The filter bank channels may be low pass filtered before downsampling.

[0028] The preprocessor may be configured to extract features of the at least one electrical input signal and / or the sensor signal. The features of the at least one electrical input signal may include, for example, modulation, level, signal-to-noise ratio (SNR), correlation, etc. The feature vector may exhibit a reduced complexity compared to the at least one electrical input signal and / or the sensor signal from which the features are extracted. The feature vector may be provided as a time-frequency representation of the audio signal (obtained by a filter bank or a warped filter bank). The time-frequency representation may be further processed into a magnitude response, and the magnitude response may be low-pass filtered and / or down-sampled. In the case where the hearing device comprises multiple microphones (and thus has access to multiple electrical input signals representing sound), some or all of the electrical input signals may be combined into a directional (beamformed) signal, such as a directional signal that enhances the user's own voice. The directional signal may be further enhanced by noise reduction, such as using a post-filter.

[0029] The time period of the corresponding values ​​of the at least one electrical input signal and optionally the sensor signal covered by the given feature vector is used as input to the input layer of the neural network, which includes at least one time frame of the at least one electrical input signal. The time period may include multiple time frames of the at least one electrical input signal, such as more than 3 time frames, such as more than 5 time frames, such as 2 to 50 time frames, for example corresponding to up to 0.5 to 1 s of audio, such as corresponding to one or more words.

[0030] The communication link established by the transmitter and receiver circuits of the transceiver of the hearing device may be a wireless link. The wireless link may be an inductive link based on inductive coupling between inductors of the hearing device and the other device or server. The wireless link may be based on radiated fields, for example based on Bluetooth or WLAN transceivers in the hearing device and the other device or server.

[0031] The hearing device is configured to operate in a plurality of operating modes, including an adaptive operating mode and a normal operating mode. Whether a user is allowed to enter the adaptive mode may depend on whether the hearing device is detected to be mounted on the ear. The adaptive operating mode may, for example, be an operating mode when the hearing device is connected to an auxiliary device (comprising a processor such as a smartphone, PC or laptop). The auxiliary device may, for example, be configured to run fitting software of the hearing device. The fitting software may be configured to customize (personalize) the algorithm of the hearing device to the needs of a specific user, as part of personalizing the hearing device to the specific user in question, including determining optimized node parameters of a neural network to be applied in the hearing device. The set of optimized node parameters of the neural network for a given user may be selected from a plurality of predetermined, optimized node parameters according to an appropriate selection algorithm. A plurality of predetermined, optimized node parameters of candidate neural networks used when customizing the hearing device may be stored in the hearing device. Thus, an appropriate set of optimized node parameters may be selected, for example, by the user himself without the use of an auxiliary device. We can also imagine decisions based on a neural network (NN) operating in one hearing device and another NN operating in another device. For example, one NN may be used in a low SNR environment, while another NN may be used in a high SNR environment.

[0032] The detector implemented by the neural network or a part thereof may be a self-voice detector and / or a keyword detector. The self-voice detector may be implemented by a neural network. A keyword detector (such as a wake-up word detector) may be implemented by a neural network. A keyword detector (such as a wake-up word detector) that is specifically optimized to detect words spoken by a user may be implemented by a neural network. Additionally or alternatively, a neural network may implement an on / off detector or a preferred speaker detector such as a spouse (voice) detector.

[0033] The hearing device may consist of or include a hearing aid, a headset, a microphone, an ear protector, or a combination thereof.

[0034] The hearing device may be adapted to provide frequency dependent gain and / or level dependent compression and / or frequency transposition of one or more frequency ranges to one or more other frequency ranges (with or without frequency compression) to compensate for a hearing impairment of the user. The hearing device may comprise a signal processor for enhancing an input signal and providing a processed output signal.

[0035] The hearing device may comprise an output unit for providing stimulation perceived by the user as an acoustic signal based on the processed electrical signal. The output unit may comprise a plurality of electrodes of a cochlear implant (for a CI type hearing device) or a vibrator of a bone conduction hearing device. The output unit may comprise an output transducer. The output transducer may comprise a receiver (speaker) for providing the stimulation as an acoustic signal to the user (e.g. in an acoustic (air conduction based) hearing device). The output transducer may comprise a vibrator for providing the stimulation as a mechanical vibration of the skull to the user (e.g. in a bone attached or bone anchored hearing device).

[0036] The hearing device comprises an input transducer for providing an electrical input signal representing a sound. The input transducer may comprise a microphone for converting the input sound into the electrical input signal. The input transducer may comprise a wireless receiver for receiving a wireless signal comprising or representing the sound and providing the electrical input signal representing the sound. The wireless receiver may, for example, be configured to receive electromagnetic signals in the radio frequency range (3kHz to 300GHz). The wireless receiver may, for example, be configured to receive electromagnetic signals in the optical frequency range (e.g. infrared light 300GHz to 430THz, or visible light, e.g. 430THz to 770THz).

[0037] A hearing device may comprise a directional microphone system adapted to spatially filter sounds from the environment so as to enhance a target sound source among a plurality of sound sources in the local environment of a user wearing the hearing device. The directional system is adapted to detect (e.g. adaptively detect) from which direction a particular portion of the microphone signal originates. This may be achieved in a number of different ways, such as described in the prior art. In hearing devices, microphone array beamformers are often used to spatially attenuate background noise sources. Many beamformer variants can be found in the literature. Minimum Variance Distortionless Response (MVDR) beamformers are widely used in microphone array signal processing. Ideally, an MVDR beamformer keeps the signal from the target direction (also called the line of sight) unchanged, while maximally attenuating sound signals from other directions. A Generalized Sidelobe Canceller (GSC) structure is an equivalent representation of an MVDR beamformer, which provides computational and digital representation advantages over a direct implementation in its original form. A hearing device may also comprise a spatially based postfilter, for example as part of a noise reduction system.

[0038] The hearing device may comprise an antenna and a transceiver circuit (e.g. a wireless receiver) for receiving a direct electrical input signal from another device, such as from an entertainment device (e.g. a television), a communication device, a wireless microphone or another hearing device. The direct electrical input signal may represent or include an audio signal and / or a control signal and / or an information signal. The hearing device may comprise a demodulation circuit for demodulating the received direct electrical input to provide a direct electrical input signal representing an audio signal and / or a control signal, for example for setting an operating parameter (e.g. volume) and / or a processing parameter of the hearing device. In general, the wireless link established by the antenna and the transceiver circuit of the hearing device may be of any type. The wireless link is established between two devices, such as between an entertainment device (e.g. a TV) and a hearing device, or between two hearing devices, for example via a third intermediate device (e.g. a processing device, such as a remote control device, a smartphone, etc.). The wireless link is typically used under power constraints, for example because the hearing device is or comprises a portable (usually battery-powered) device. The wireless link is a link based on near field communication, such as an inductive link based on inductive coupling between antenna coils of the transmitter part and the receiver part. In another embodiment, the wireless link is based on far field electromagnetic radiation. The communication via the wireless link is arranged according to a specific modulation scheme, such as an analog modulation scheme, such as FM (frequency modulation) or AM (amplitude modulation) or PM (phase modulation), or a digital modulation scheme, such as ASK (amplitude shift keying) such as on-off keying, FSK (frequency shift keying), PSK (phase shift keying) such as MSK (minimum shift keying) or QAM (quadrature amplitude modulation), etc.

[0039] The communication between the hearing device and the other device is in baseband (audio frequency range, e.g. between 0 and 20 kHz). Preferably, the frequency used for establishing the communication link between the hearing device and the other device is below 70 GHz, e.g. in the range from 50 MHz to 70 GHz, e.g. above 300 MHz, e.g. in the ISM range above 300 MHz, e.g. in the 900 MHz range or in the 2.4 GHz range or in the 5.8 GHz range or in the 60 GHz range (ISM = Industrial, Scientific and Medical, such standardized ranges are defined, e.g., by the International Telecommunication Union ITU). The wireless link is based on a standardized or proprietary technology. The wireless link is based on Bluetooth technology (e.g. Bluetooth low power technology).

[0040] The hearing device may be or form part of a portable (ie configured to be wearable) device, such as a device comprising a local energy source such as a battery, such as a rechargeable battery. The hearing device may for example be a lightweight, easily wearable device, such as having a total weight of less than 100g, such as less than 20g.

[0041] The hearing device may comprise a forward or signal path between an input unit, such as an input transducer, for example a microphone or a microphone system and / or a direct electrical input, such as a wireless receiver, and an output unit, such as an output transducer. A signal processor is located in the forward path. The signal processor is adapted to provide a frequency-dependent gain according to the specific needs of the user. The hearing device may comprise an analysis path having functional parts for analyzing the input signal, such as determining the level, modulation, signal type, acoustic feedback estimate, etc. Part or all of the signal processing in the analysis path and / or the signal path is performed in the frequency domain. Part or all of the signal processing in the analysis path and / or the signal path is performed in the time domain.

[0042] The analog electrical signal representing the acoustic signal is converted into a digital audio signal in an analog-to-digital (AD) conversion process, in which the analog signal is sampled at a predetermined sampling frequency or sampling rate f. s Sampling, f s For example, in the range from 8 kHz to 48 kHz (adapted to the specific needs of the application) at discrete time points t n (or n) provides digital samples x n (or x[n]), each audio sample is passed through a predetermined N b The bit represents the sound signal at t n The value of N b For example, in the range from 1 to 48 bits, such as 24 bits. Each audio sample thus uses N b bit quantization (resulting in 2 Nb different possible values). A digital sample x has a 1 / f s The length of time for f s = 20kHz, such as 50μs. Multiple audio samples are arranged in time frames. A time frame includes 64 or 128 audio data samples. Other frame lengths can be used according to actual applications.

[0043] The hearing device may comprise an analogue-to-digital (AD) converter to digitise an analogue input (e.g. from an input transducer such as a microphone) at a predetermined sampling rate, e.g. 20 kHz. It should be mentioned that, although the (maximum) sampling rate in the hearing device is e.g. 20 kHz in the forward path of the hearing device (supporting an audio frequency range of 0-10 kHz), a different, smaller frequency range may be used in other parts of the hearing device. A digital-to-analogue (DA) converter is included to convert the digital signal into an analogue output signal, e.g. for presentation to a user via an output transducer.

[0044] The hearing device, such as an input transducer and / or an antenna and transceiver circuitry, comprises a TF conversion unit for providing a time-frequency representation of an input signal. The time-frequency representation may comprise an array or a mapping of corresponding complex or real values ​​of the signal concerned in a certain time and frequency range. The TF conversion unit may comprise a filter bank for filtering the (time-varying) input signal and providing a plurality of (time-varying) output signals, each output signal comprising a distinct frequency range of the input signal. The TF conversion unit may comprise a Fourier transform unit for converting the time-varying input signal into a (time-varying) signal in the (time-)frequency domain. The minimum frequency f is taken into account by the hearing device. min To the maximum frequency f max The frequency range of includes a portion of the typical human hearing range from 20 Hz to 20 kHz, for example a portion of the range from 20 Hz to 12 kHz. Typically, the sampling rate f s Greater than or equal to the maximum frequency f max twice, that is, f s ≥2f max The signal of the forward path and / or the analysis path of the hearing device is split into NI frequency bands (e.g. of uniform width), wherein NI is e.g. greater than 5, such as greater than 10, such as greater than 50, such as greater than 100, such as greater than 500, at least parts of which are processed separately. The hearing device is adapted to process the signal of the forward and / or analysis path at NP different frequency channels (NP≤NI). For a detector, e.g. a detector for analyzing a signal of the forward path, such as an electrical input signal, we may have a smaller number of frequency channels, e.g. NP'≤NP. The frequency channels may be of uniform or non-uniform width (e.g. the width increases with frequency), overlapping or non-overlapping.

[0045] The hearing device may be configured to operate in different modes, such as a normal mode and one or more specific modes, e.g. selectable by a user or selectable automatically. The operating modes may be optimized for specific acoustic situations or environments. The operating modes may include a low power mode, in which the functionality of the hearing device is reduced (e.g. in order to save energy), such as disabling wireless communication and / or disabling specific features of the hearing device. The operating modes may include a specific adaptation mode, in which the hearing device is connected to a plurality of different candidate neural networks (e.g. located in the hearing device or in an auxiliary device), one of the candidate neural networks being intended to be selected for use in the hearing device.

[0046] The hearing device may comprise a plurality of detectors configured to provide status signals relating to a current network environment (e.g. a current acoustic environment) of the hearing device, and / or relating to a current state of a user wearing the hearing device, and / or relating to a current state or operating mode of the hearing device. Alternatively or additionally, one or more of the detectors may form part of an external device that communicates with the hearing device (e.g. wirelessly). The external device may, for example, comprise another hearing device, a remote control, an audio transmission device, a phone (e.g. a smartphone), an external sensor, etc.

[0047] One or more of the plurality of detectors may be configured to operate on full-band signals (time domain). One or more of the plurality of detectors may be configured to operate on band-split signals ((time-)frequency domain), for example in a limited plurality of frequency bands.

[0048] The plurality of detectors may include a level detector for estimating the current level of the signal of the forward path. The predetermined criterion may include whether the current level of the signal of the forward path is above or below a given (L-)threshold. One or more level detectors may act on a full-band signal (time domain). One or more level detectors may act on a band-split signal ((time-)frequency domain).

[0049] The hearing device may comprise a voice activity detector (VAD) for estimating whether (or with what probability) an input signal (at a specific point in time) comprises a voice signal. In the present specification, a voice signal comprises a speech signal from a human being. It may also comprise other forms of vocalizations (such as singing) produced by a human speech system. The voice activity detector unit is adapted to classify the user's current acoustic environment as a "voice" or a "no-voice" environment. This has the advantage that time periods comprising an electric microphone signal of human vocalizations (such as speech) in the user's environment can be identified and thus separated from time periods comprising only (or mainly) other sound sources (such as artificially produced noise). The voice activity detector is adapted to detect the user's own voice as "voice" as well. As an alternative, the voice detector is adapted to exclude the user's own voice from the detection of "voice".

[0050] The hearing device may comprise a self-voice detector for estimating whether (or with what probability) a particular input sound (e.g. voice, such as speech) originates from the voice of a user of the hearing system. The microphone system (or self-voice detector) of the hearing device may be adapted to be able to distinguish the user's own voice from the voice of another person and possibly from unvoiced sounds.

[0051] The plurality of detectors may include a motion detector, such as an accelerometer, configured to detect movement of the user's facial muscles and / or bones, such as due to speech or chewing (eg, jaw movement), and provide a detector signal indicative of the movement.

[0052] The hearing device may comprise a classification unit configured to classify a current situation based on an input signal from (at least part of) the detector and possibly other inputs. In this specification, a "current situation" is defined by one or more of the following:

[0053] a) the physical environment (e.g. including the current electromagnetic environment, such as the presence of electromagnetic signals (including audio and / or control signals) intended or not intended to be received by the hearing device, or other properties of the current environment other than acoustics);

[0054] b) Current acoustic conditions (input level, feedback, etc.);

[0055] c) the user’s current mode or state (motion, temperature, cognitive load, etc.);

[0056] d) The current mode or status of the hearing device and / or another device communicating with the hearing device (selected program, time elapsed since last user interaction, etc.).

[0057] The classification unit may be based on or include a neural network, such as a trained neural network.

[0058] The hearing device may also include other suitable functions for the application in question, such as compression, noise reduction, feedback control etc.

[0059] The hearing device may include a listening device, such as a hearing aid, such as a hearing instrument, such as a hearing instrument adapted to be located at the ear or completely or partially located in the ear canal of a user, such as a headset, an earphone, an ear protection device or a combination thereof. The hearing aid system may include a speaker amplifier (including multiple input transducers and multiple output transducers, such as used in audio conferencing situations), such as including a beamformer filter unit, such as providing multiple beamforming capabilities.

[0060] application

[0061] In one aspect, there is provided an application of a hearing device as described above, in detail in the "Detailed Description" section and in the claims. An application in a system comprising audio distribution may be provided. An application in a system comprising one or more hearing aids (such as hearing instruments), headphones, earphones, active ear protection systems, etc. may be provided, for example, use in a hands-free telephone system, a teleconferencing system (e.g. comprising a loudspeaker amplifier), a broadcasting system, a karaoke system, a classroom amplification system, etc.

[0062] method

[0063] In one aspect, the present application also provides a method for selecting optimized parameters of a neural network used in a portable hearing device. The method comprises:

[0064] - providing a portable hearing device to be used by a specific user, the hearing device comprising a neural network processor adapted to implement a neural network comprising an input layer, an output layer and a plurality of hidden layers, each layer comprising a plurality of nodes, each node being defined by a plurality of node parameters and a non-linear function, the neural network being configured to receive an input vector and to provide an output vector which is a specific non-linear function of the input vector;

[0065] - fitting the hearing device to the user;

[0066] -Provide auxiliary devices;

[0067] - establishing a communication link enabling the exchange of data between the hearing device and the assistive device;

[0068] In the hearing device, the method may further comprise:

[0069] - providing at least one electrical input signal representative of sounds in the environment of a user wearing the hearing device;

[0070] - processing the at least one electrical input signal and providing a feature vector representing a time segment of the at least one electrical input signal;

[0071] - transmitting said feature vector to said auxiliary device via said communication link.

[0072] In the auxiliary device, the method may further include:

[0073] - providing a plurality of pre-trained candidate neural networks, each candidate neural network having the same structure as a neural network of a hearing device, wherein each pre-trained network is considered as a candidate network for the person, and wherein each pre-trained neural network has been trained based on completely or partially different training data;

[0074] - receiving the feature vector from the hearing device and providing it as an input vector to the plurality of pre-trained candidate neural networks;

[0075] - determining corresponding output vectors corresponding to the feature vectors from the plurality of pre-trained candidate neural networks;

[0076] - comparing the output vectors and selecting one of a plurality of candidate neural networks as a neural network optimized for the hearing device according to a predetermined criterion;

[0077] - transmitting node parameters of a selected neural network from a plurality of candidate neural networks to the hearing device via the communication link, and

[0078] In the hearing device, the method may further comprise:

[0079] - receiving the node parameters and feeding them to the neural network processor and applying them to the neural network.

[0080] In another aspect, the present invention provides a method for selecting optimized parameters of a neural network for use in a portable hearing device. The method comprises:

[0081] - providing a portable hearing device to be used by a specific user, the hearing device comprising a neural network processor adapted to implement a neural network comprising an input layer, an output layer and a plurality of hidden layers, each layer comprising a plurality of nodes, each node being defined by a plurality of node parameters and a non-linear function, the neural network being configured to receive an input vector and to provide an output vector which is a specific non-linear function of the input vector;

[0082] - fitting the hearing device to the user;

[0083] - providing at least one electrical input signal representative of sounds in the environment of a user wearing the hearing device;

[0084] - processing the at least one electrical input signal and providing a plurality of feature vectors, each feature vector representing a time period of the at least one electrical input signal;

[0085] - providing a plurality of pre-trained candidate neural networks, wherein each pre-trained network is considered as a candidate network of the user, and wherein each pre-trained neural network has been trained based on completely or partially different training data;

[0086] - receiving the feature vector and providing it as an input vector to the plurality of pre-trained candidate neural networks;

[0087] - determining corresponding output vectors corresponding to the feature vectors from the plurality of pre-trained candidate neural networks;

[0088] - comparing the output vectors and selecting one of a plurality of candidate neural networks as a neural network optimized for the hearing device based on a predetermined criterion on the output vectors;

[0089] - transmitting node parameters of a selected neural network from a plurality of candidate neural networks to a neural network processor of the hearing device, and

[0090] - receiving the node parameters into the neural network processor and applying them to the neural network.

[0091] When appropriately replaced by corresponding processes, part or all of the structural features of the apparatus described above, described in detail in the "Detailed Description" or defined in the claims may be combined with the implementation of the corresponding method, and vice versa. The implementation of the method has the same advantages as the corresponding apparatus.

[0092] The term "neural network having the same structure as the neural network of the hearing device" means that each candidate neural network has the same number of layers and each layer has the same number of nodes, but each candidate neural network has different node parameters (weights and / or biases and / or nonlinear functions) determined by pre-training with different training data sets. A given neural network can be defined by a given set of node parameters.

[0093] Different candidate neural networks may, for example, be trained based on data from different groups of people (differentiated by age, gender), with the network selected based on providing information about the person (such as age and / or gender).

[0094] The term "pre-trained neural network has been trained based on completely or partially different training data" includes, for example, that the noise portion of the training data may be similar. The noise may, for example, include training data that will not detect any relevant speech elements. The training data may also include speech data that is close to (but different from) words that are expected to be detected by a detector (such as a keyword detector).

[0095] The predetermined criteria may be based on the performance of the neural network in terms of true positives, false positives, true rejections and false rejections.

[0096] From the four terms TP, TN, FP and FN (True Positive, True Negative, False Positive, False Negative), different metrics can be derived, such as "Accuracy" = TP+TN / (TP+TN+FP+FN), "Precision" = TP / (TP+FP), or "Recall" = TP / (TP+FN) (see, for example https: / / developers.google.com / machine-learning / crash-course / classification / accuracy ? ).

[0097] Each candidate neural network may have the same structure. At least some (e.g., one) of the candidate neural networks may have a different structure from the other candidate neural networks, such as a different number of layers or nodes. The candidate neural networks may include different types of neural networks. The type of neural network may be selected from, for example, feedforward, recursive, convolutional, etc.

[0098] Each candidate neural network may have been trained based on training data from different categories of people. Such a training procedure may be suitable for self-voice applications such as self-voice detection (OVD) and keyword spotting (KWS) (e.g., wake-up word detection). Different groups of people belong to different categories. The categories may be derived from, for example, combining Figure 6 Overview of basic population generation. Different categories of people may exhibit different acoustic properties, such as different head-related transfer functions (HRTFs), different voices ("spectral signatures", fundamental and / or format frequencies, etc.), different ages, different genders, etc.

[0099] A signal (eg from a sensor) representing a current value of a characteristic of the user or the user's environment may be provided in the hearing device together with the at least one electrical input signal and processed to provide the feature vector.

[0100] The method of the present invention may be configured such that the neural network implements an own voice detector (OVD) and / or a keyword detector (KWD).

[0101] The method may include prompting the user to speak. The method may be configured to prompt the user to speak one or more (e.g. predetermined) words or sentences (e.g. including words or sentences similar to the words or sentences that the detector is intended to recognize or detect). Thus, the one or more words or sentences may form the basis of at least part of the feature vectors in the plurality of feature vectors. The output vectors (ground truth) of the candidate neural networks are thus known, which can be used to check which candidate neural networks (i.e. each of their neural network parameters, e.g. parameters optimized for a particular class of people) meet the criteria (e.g. best suited for the current user).

[0102] The predetermined criteria for selecting the optimized neural network parameters may be based on a prompt word or sentence spoken by the user. The optimized neural network parameters may be selected by comparing output vectors from candidate neural networks having corresponding neural network parameters (e.g., parameters optimized for a particular class of people) based on the prompt word or sentence spoken by the user. The selected neural network parameters may be parameters from the candidate neural network that best meets the predetermined criteria (e.g., has the highest number of correct values ​​for the output vector when the user speaks the prompt word or sentence).

[0103] The method of the invention may comprise providing several groups of a plurality of pre-trained candidate neural networks, each candidate neural network of each group having the same structure as the neural network of the hearing device, wherein each pre-trained network is regarded as a candidate network of the person, and wherein each pre-trained neural network has been trained based on completely or partially different training data, and wherein each group of pre-trained candidate neural networks aims to implement a different detector. Thus, the selection of optimized parameters of several different neural networks implementing different detectors may be performed simultaneously. The different detectors may, for example, comprise a keyword detector for detecting a limited number of keywords (such as command words), a wake-up word detector for detecting a specific word or word combination for activating a voice interface, and so on.

[0104] Computer readable medium or data carrier

[0105] The present invention further provides a tangible computer-readable medium (or data carrier) storing a computer program including program code (instructions), which, when executed on a data processing system, enables the data processing system (computer) to execute (complete) at least part (such as most or all) of the steps of the method described above, described in detail in the "Specific Implementation Method" and defined in the claims.

[0106] As an example but not limitation, the aforementioned tangible computer readable medium may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage device, or any other medium that can be used to execute or store the desired program code in the form of instructions or data structures and can be accessed by a computer. As used herein, disks include compact disks (CDs), laser disks, optical disks, digital versatile disks (DVDs), floppy disks, and blue-ray disks, wherein these disks usually reproduce data magnetically, while these disks can reproduce data optically with lasers. Other storage media include storage in DNA (e.g., in synthetic DNA chains). Combinations of the above disks should also be included in the scope of computer readable media. In addition to being stored on tangible media, computer programs can also be transmitted via transmission media such as wired or wireless links or networks such as the Internet and loaded into a data processing system to run at a location different from the tangible media.

[0107] Computer Programs

[0108] In addition, the present application provides a computer program (product) comprising instructions, which, when executed by a computer, causes the computer to execute (the steps of) the method described above, described in detail in the “Detailed Description” and defined in the claims.

[0109] Data processing system

[0110] On the one hand, the present invention further provides a data processing system comprising a processor and a program code, wherein the program code enables the processor to perform at least part (such as most or all) of the steps of the method described above, described in detail in the "Specific Implementation Method" and defined in the claims.

[0111] Hearing System

[0112] In another aspect, there is provided a hearing device and a hearing system including an auxiliary device as described above, in detail in the “Detailed Description of the Invention” and in the claims.

[0113] The hearing system is adapted to establish a communication link between the hearing device and the auxiliary device so that information (such as control and status signals, possibly audio signals) can be exchanged or forwarded from one device to the other.

[0114] Auxiliary devices may include remote controls, smart phones, or other portable or wearable electronic devices such as smart watches.

[0115] The auxiliary device may be or include a remote control for controlling the functions and operation of the hearing device. The functions of the remote control are implemented in a smartphone, which may run an APP that enables control of the functions of the audio processing device via the smartphone (the hearing device includes a suitable wireless interface to the smartphone, e.g. based on Bluetooth or some other standardized or proprietary solution).

[0116] The auxiliary device may be or include an audio gateway device adapted to receive a plurality of audio signals (e.g. from an entertainment device such as a TV or music player, from a telephone device such as a mobile phone or from a computer such as a PC) and to select and / or combine an appropriate signal (or combination of signals) of the received audio signals for transmission to the hearing device.

[0117] The auxiliary device may do multiple parts of the processing (the division may be, for example, that the OVD and several keywords related to the functionality of the hearing device (including a wake-up word for a voice control interface) are detected in the hearing device, while other keywords are detected in the auxiliary device).

[0118] The auxiliary device may consist of or may comprise another hearing device.The hearing system may comprise two hearing devices adapted to implement a binaural hearing system, such as a binaural hearing aid system.

[0119] APP

[0120] On the other hand, the present invention also provides a non-transient application called APP. The APP may include executable instructions configured to be run on an auxiliary device to implement a user interface for the hearing device or hearing system described above, described in detail in the "Detailed Description of the Invention" and defined in the claims. The APP is configured to be run on a mobile phone such as a smart phone or another portable device that enables communication with the hearing device or hearing system.

[0121] definition

[0122] In this specification, a "hearing device" refers to a device such as a hearing aid, for example a hearing instrument or an active ear protection device or other audio processing device, which is suitable for improving, enhancing and / or protecting the hearing ability of a user by receiving an acoustic signal from the user's environment, generating a corresponding audio signal, possibly modifying the audio signal, and providing the possibly modified audio signal as an audible signal to at least one ear of the user. A "hearing device" also refers to a device such as a headset or an earphone, which is suitable for electronically receiving an audio signal, possibly modifying the audio signal, and providing the possibly modified audio signal as an audible signal to at least one ear of the user. The audible signal can be provided, for example, in the form of an acoustic signal radiated into the outer ear of the user, an acoustic signal transmitted to the inner ear of the user as a mechanical vibration through the bone structure of the user's head and / or through parts of the middle ear, and an electrical signal transmitted directly or indirectly to the cochlear nerve of the user.

[0123] The hearing device may be configured to be worn in any known manner, such as as a unit worn behind the ear (with a tube directing the radiated acoustic signal into the ear canal or with an output transducer such as a loudspeaker arranged close to or in the ear canal), as a unit arranged wholly or partly in the pinna and / or ear canal, as a unit connected to a fixed structure implanted in the skull such as a vibrator, or as a connectable or wholly or partly implanted unit, etc. The hearing device may comprise a single unit or several units in electronic communication with each other. The loudspeaker may be arranged in a housing together with the other components of the hearing device, or it may be an external unit itself (possibly in combination with a flexible guiding element such as a dome-shaped element).

[0124] More generally, a hearing device comprises an input transducer for receiving an acoustic signal from the user's environment and providing a corresponding input audio signal and / or a receiver for receiving the input audio signal electronically (i.e. wired or wirelessly), a (typically configurable) signal processing circuit (e.g. a signal processor, e.g. comprising a configurable (programmable) processor, e.g. a digital signal processor) for processing the input audio signal, and an output unit for providing an audible signal to the user based on the processed audio signal. The signal processor may be adapted to process the input signal in the time domain or in a plurality of frequency bands. In some hearing devices, an amplifier and / or a compressor may constitute the signal processing circuit. The signal processing circuit typically comprises one or more (integrated or separate) memory elements for executing a program and / or for storing parameters used (or likely to be used) in the processing and / or for storing information suitable for the functioning of the hearing device and / or for storing information (e.g. processed information, e.g. provided by the signal processing circuit) for use, e.g. in conjunction with an interface to a user and / or to an interface to a programming device. In some hearing devices, the output unit may comprise an output transducer, e.g. a loudspeaker for providing an air-borne acoustic signal or a vibrator for providing a structure- or liquid-borne acoustic signal. In some hearing devices, the output unit may include one or more output electrodes for providing an electrical signal (e.g. a multi-electrode array for electrical stimulation of the cochlear nerve). The hearing device may include a loudspeaker amplifier (comprising multiple input transducers and multiple output transducers, e.g. for use in audio conferencing situations).

[0125] In some hearing devices, the vibrator may be adapted to transmit structure-borne acoustic signals to the skull percutaneously or by the skin. In some hearing devices, the vibrator may be implanted in the middle ear and / or the inner ear. In some hearing devices, the vibrator may be adapted to provide structure-borne acoustic signals to the bones of the middle ear and / or the cochlea. In some hearing devices, the vibrator may be adapted to provide liquid-borne acoustic signals to the cochlear fluid, for example, through the oval window. In some hearing devices, the output electrode may be implanted in the cochlea or on the inside of the skull and may be adapted to provide electrical signals to the hair cells of the cochlea, one or more auditory nerves, the auditory brainstem, the auditory midbrain, the auditory cortex, and / or other parts of the cerebral cortex.

[0126] A hearing device such as a hearing aid may be adapted to the needs of a specific user such as hearing loss. The configurable signal processing circuit of the hearing device may be adapted to apply frequency- and level-dependent compression amplification of an input signal. The customized frequency- and level-dependent gain (amplification or compression) may be determined during the fitting process by the fitting system based on the user's hearing data such as an audiogram using the basic principles of fitting (e.g. adaptation to speech). The frequency- and level-dependent gain may, for example, be embodied in a processing parameter, uploaded to the hearing device, for example via an interface to a programming device (fitting system), and used by a processing algorithm executed by the configurable signal processing circuit of the hearing device.

[0127] "Hearing system" refers to a system including one or two hearing devices. "Binaural hearing system" refers to a system including two hearing devices and adapted to provide audible signals to the two ears of a user in a coordinated manner. A hearing system or binaural hearing system may also include one or more "auxiliary devices" that communicate with the hearing device and affect and / or benefit from the function of the hearing device. An auxiliary device may be, for example, a remote controller, an audio gateway device, a mobile phone (such as a smart phone) or a music player. A hearing device, a hearing system or a binaural hearing system may be used, for example, to compensate for the hearing loss of a hearing-impaired person, enhance or protect the hearing ability of a normal hearing person, and / or transmit an electronic audio signal to a person. A hearing device or a hearing system may, for example, form part of a broadcasting system, an active ear protection system, a hands-free telephone system, a car audio system, an entertainment (such as karaoke) system, a teleconferencing system, a classroom amplification system, or interact with it.

[0128] The invention may be used, for example, in applications such as hearing aids, headphones or similar devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0129] Various aspects of the present invention will be best understood from the detailed description below in conjunction with the accompanying drawings. For clarity, the drawings are schematic and simplified, and only the details necessary for understanding the present invention are given, while other details are omitted. Throughout the specification, the same reference numerals are used for the same or corresponding parts. The various features of each aspect may be combined with any or all features of the other aspects. These and other aspects, features and / or technical effects will be apparent from and illustrated in conjunction with the following figures, in which:

[0130] Figure 1 shows a part of a hearing instrument with a built-in detector according to the invention;

[0131] Figure 2 shows a hearing instrument wirelessly connected to an external device, which illustrates the selection procedure of an optimized neural network used in the hearing instrument according to the present invention;

[0132] Figure 3 An exemplary personalization procedure of hearing device parameters according to the present invention is shown;

[0133] Figure 4 A flow chart showing an embodiment of the proposed method for selecting a personalized neural network;

[0134] Figure 5 A hearing device according to an embodiment of the invention is shown, which uses a trained (personalized for a specific user) neural network to control processing of a signal representing sound in the hearing device before the processed signal is presented to a user wearing the hearing device;

[0135] Figure 6 An exemplary procedure for subdividing a basic population for providing training data for training a neural network into a plurality of subgroups for training a plurality of neural networks to provide a plurality of optimized neural networks, each optimized neural network representing a different characteristic of a subject;

[0136] Fig. 7A An embodiment of a keyword detector according to the present invention implemented as a neural network is shown;

[0137] Figure 7B shows the inclusion of Fig. 7A The neural network input vector is the context of the audio electrical input signal;

[0138] Figure 8 An embodiment of a hearing device according to the invention is shown, comprising an adaptation unit configured to enable selection of a set of optimization parameters for a neural network among a plurality of optimization parameters without using external equipment.

[0139] By the detailed description given below, the further scope of application of the present invention will be apparent. However, it should be understood that while the detailed description and specific examples show the preferred embodiments of the present invention, they are only provided for illustrative purposes. For those skilled in the art, based on the following detailed description, other embodiments of the present invention will be apparent. DETAILED DESCRIPTION

[0140] The detailed description proposed below in conjunction with the accompanying drawings serves as a description of a variety of different configurations. The detailed description includes specific details for providing a thorough understanding of a number of different concepts. However, it is apparent to those skilled in the art that these concepts can be implemented without these specific details. Several aspects of the apparatus and method are described by a number of different blocks, functional units, modules, components, circuits, steps, processes, algorithms, etc. (collectively referred to as "elements"). Depending on the specific application, design limitations or other reasons, these elements can be implemented using electronic hardware, computer programs or any combination thereof.

[0141] The electronic hardware may include microprocessors, microcontrollers, digital signal processors (DSPs), field programmable gate arrays (FPGAs), programmable logic devices (PLDs), gating logic, discrete hardware circuits, and other appropriate hardware configured to perform a number of different functions described in this specification. Computer programs should be broadly interpreted as instructions, instruction sets, codes, code segments, program codes, programs, subroutines, software modules, applications, software applications, software packages, routines, subroutines, objects, executables, execution threads, programs, functions, etc., whether referred to as software, firmware, middleware, microcode, hardware description language, or other names.

[0142] The present application relates to the field of hearing devices such as hearing aids. Figure 1 A hearing instrument system as shown in , with a microphone and possibly other sensors such as an accelerometer, a magnetometer, an EEG sensor, and / or a heart rate detector, etc. The hearing instrument may have a built-in detector (eg including one or more of the mentioned sensors).

[0143] One option could be to keep only some of the layers, for example keeping the weights of the first layer (or first few layers) of the neural network fixed and only updating the deeper layers. This way fewer parameters need to be programmed.

[0144] One way to achieve better performance is to personalize the parameters of the neural network / detector. For example, a neural network optimized for a specific person or a specific population (such as male, female or child voices, different ages, different languages ​​or noise environments common to a given person) may work better than a neural network that has been optimized to work well for any population.

[0145] Here we propose a personalized neural network selection method.

[0146] Figure 1 A part of a hearing instrument with a built-in detector is shown. Input signals (IN1, IN2, SIN) from one or more microphones (M1, M2) and / or one or more sensors SENSE are preprocessed (see unit Pre-PRO) into a feature vector FV. One or more sensors SENSE may be omitted, so that the input signal of the preprocessor Pre-PRO is only the electrical input signal (IN1, IN2) from the microphones (M1, M2), so that the feature vector depends only on the electrical input signal (IN1, IN2) from the microphones (M1, M2). The preprocessed feature vector FV is used as input to a neural network NN. The output of the neural network NN may be, for example, a probability or a set of probabilities p(x) for making a decision and / or a detection (e.g. detection of a specific word, detection of one's own voice or detection of a certain sound environment, see decision unit PostPRO and output RES). The output RES of the decision unit may be used, for example, to decide on a specific action in the hearing device, such as activating a voice control interface. The output RES of the decision unit may be transmitted to another device or system, for example, for activating a service.

[0147] exist Figure 1In the embodiment, the hearing device comprises a single neural network implementing a (single) detector, such as a self-voice detector. The hearing device may, for example, comprise several neural networks operating in parallel. The hearing device may, for example, comprise one neural network for implementing a keyword detector and another neural network for implementing a self-voice detector. In this case, the scheme for selecting an optimized neural network for a specific user among a plurality of optimized candidate neural networks is the same, and may even be optimized simultaneously, for example. In this case, there are two or more groups, such as a first and a second group, respectively, of a plurality of optimized candidate networks, each group of neural networks being optimized to implement a specific detector (e.g. a keyword detector and a self-voice detector, respectively), each candidate network in the first and the second group simultaneously receiving the same input vector from the user (from the hearing device) (see Figure 2 , for a single set of candidate networks). When referring to multiple sets of networks representing different detectors, the first and second sets do not necessarily have the same input vectors. A set of (candidate) NNs (each NN implementing a given type of OV detector) may have different input features than the set of NNs used for wake-up word detection.

[0148] A neural network can be viewed as a generalized non-linear function of an input signal optimized to achieve an output signal of a certain property. While passing a signal through a neural network has a relatively low complexity, estimating the parameters of the neural network is computationally very heavy and (very) time consuming. To personalize a neural network, the neural network needs to be trained based on data from a specific person. Even if the network can be only partially trained for an individual, for example using known transfer learning techniques, the training process and data collection is still very time consuming. With regard to transfer learning, the hearing device can be configured to update only a portion of the neural network parameters (e.g. the parameters of the p last layers). Thereby, fewer parameters in the hearing instrument need to be updated.

[0149] As an alternative to optimizing a neural network for an individual, we propose to select among multiple pre-trained neural networks, where each pre-trained network is considered a candidate network for that person. Each pre-trained neural network can be trained on data from a certain class of people. The number of pre-trained neural networks depends on how the actual classification of population classes is to be performed.

[0150] An exemplary neural network including input and output vectors is shown in Fig. 7A , 7B However, it should be emphasized that other NN structures may also be used.

[0151] consider Figure 2 The system shown in . Figure 2A hearing instrument is shown wirelessly connected to an external device, which illustrates the selection procedure of an optimized neural network used in the hearing instrument according to the invention. While the hearing instrument HD (due to size restrictions) has limited memory and computing power, the external device ExD has much more computing power and available memory. This enables the external device to quickly evaluate the (possibly pre-processed) hearing instrument sound (in the form of feature vectors FV), which are provided as several neural networks (NN1, ..., NN K ) input FV', these neural networks have for example been pre-trained for different populations but implement the same detector, such as a self-speech detector. Based on a sound example FV' from a hearing device HD, see for example Figure 3 , evaluate different neural networks (NN1,…,NN K ). The parameters of the network with the best performance are used in the hearing instrument. An evaluation of the relative performance of the neural network can be made, for example, in terms of a comparison of the number of true positives, false positives, true rejections and false rejections. Figure 2 In the example, the outputs of different neural networks are the corresponding probabilities p of self-voice detection. i (OV), i = 1, ..., K. These probabilities are compared, for example, with the true value (i.e., whether the feature vector represents a sound spoken by the user) to derive the network that performs best for a given person. In another example, the neural network may be optimized to detect Q predetermined keywords, such as a voice interface. In this case, the output vector of each neural network will represent the probability that a given input feature vector (e.g., derived from a word spoken by the user and picked up by the hearing aid microphone) is equal to each predetermined keyword (the output vector includes p(KWi), i = 1, ..., Q). The number of keywords Q may, for example, be in the range between 1 and 10. A wake-up word detector (Q = 1) may be implemented to detect a single keyword or key phrase, such as "Hey Siri" or "Open sesame", etc.

[0152] The system consists of a hearing device HD capable of wirelessly transmitting an audio signal or a feature vector FV derived from the audio signal (via a wireless link WLNK such as an audio link) to an external device ExD. The external device ExD (such as a smartphone or PC used during fitting of the hearing device) has much more memory and computing power than the hearing instrument and is capable of evaluating different pre-trained neural networks for neural network parameters to select (among candidate networks) a set of parameters that best suits the hearing device user. Based on different audio examples, the external device can select the best candidate network, and the hearing device will be updated with the parameters of the best candidate network.

[0153] To save computation and transmit as little data as possible, the audio recorded at the hearing device, possibly together with sensor data such as data obtained from an accelerometer, can be preprocessed (see unit Pre-PRO) into a feature vector FV, which is used as input to the neural network. An example of a feature vector can be a time-frequency representation of the audio signal (obtained by a filter bank or a warped filter bank). The time-frequency representation can be further processed into a magnitude response, and the magnitude response can be low-pass filtered and / or downsampled. Different microphone signals can be combined into a directional microphone signal, for example a directional signal that enhances the own voice. The directional signal can be further enhanced by noise reduction using a postfilter.

[0154] In the case of relatively low complexity, the hearing device may be configured to retain candidate neural networks (node ​​parameters optimized for which may be stored in the hearing device before fitting for a specific user). Thus, the selection and installation of the candidate neural network that best suits the needs of the user may be performed entirely by the hearing device itself (the hearing device does not have to be connected to an external device, and transceiver circuitry may be omitted).

[0155] The feature vector FV may depend on the application. The feature vector FV may for example be or include the complex valued output from the filter bank, or simply the magnitude (or the square of the magnitude) of the filter bank output. Alternatively or additionally the feature vector may be cepstral coefficients such as Mel Frequency Cepstral Coefficients (MFCC) or Bark Frequency Cepstral Coefficients (BFCC). In the case of self-voice detection, the feature vector FV may contain information about the transfer function between different microphone signals.

[0156] Figure 3 An exemplary personalization procedure of hearing device parameters according to the present invention is shown.

[0157] Figure 3 An example of how the proposed training procedure can be used is shown. Via an external device ExD, the user is prompted to repeat an audio sequence such as a specific word (e.g., for keyword detection, the audio sequence may consist of keywords and confusion words). The audio sequence, here the word "cheese" (or features derived therefrom) is picked up by the hearing device (HD1, HD2) and passed to the external device ExD (possibly as a pre-processed feature vector). The data is applied to K different pre-trained neural networks (NN1, ..., NN2). K). For each prompted input, the external device evaluates each candidate neural network, such as the probability of a correct word (here p(cheese)), the probability of a false detection (here p(tease)), a receiver operating curve or other performance metric. Based on the evaluated words (e.g., all prompted words), the best performing network is selected and the parameters of this network are programmed into the hearing instrument. For evaluation in different noisy environments, different types of background noise can be added to the sound, such as recorded in the external device. Figure 4 A flow chart showing a method for selecting optimization parameters of a neural network.

[0158] Figure 4 A flow chart showing an embodiment of the proposed method for selecting a personalized neural network. The method comprises the steps of:

[0159] S1: Start the personalization process;

[0160] S2: prompt word;

[0161] S3: transmitting the words and words / features spoken by the user to an external device;

[0162] S4: Calculate and update each pre-trained neural network (see Figure 2 , 3 NN1,…,NN in K ) performance;

[0163] S5: End? If no, return to step S2; if yes, proceed to the next step;

[0164] S6: Updating the hearing device with the parameters of the best performing network (see Passing the parameters of the neural network NNx to the hearing device and applying Figure 2 The neural network NN* in .

[0165] The user may also be prompted for other words, such as typical confusion words, such as peace-cheese, or prompted to read text that does not contain the desired word.

[0166] In the case of own voice (OV) detection, one can imagine different networks trained for multiple groups with similar OV transfer functions between microphones. Given an OV transfer function (TRF) measured for an individual, the difference between the measured OV TRF and the OV TRF representing each neural network can be measured. The neural network represented by the OV TRF with the highest similarity can then be selected for use in the hearing device. Alternatively, the similarity between the measured OV TRF and the OV TRFs representing different neural networks can be measured based on the neural network that provides the best detection of OV.

[0167] Figure 5A hearing device according to an embodiment of the invention is shown, which uses a trained (personalized for a specific user) neural network to control the processing of a signal representing sound before the processed signal is presented to a user wearing the hearing device. The hearing device HD comprises an input-detector-decision module such as Figure 1 As shown in . The output RES of the decision unit Post-Pro is fed to the processor PRO of the hearing device. The processor PRO receives electrical input signals (IN1, IN2) from the microphones (M1, M2) and processes these signals according to the output RES of the decision unit Post-Pro. The output RES of the decision unit Post-Pro may, for example, represent a self-voice detection control signal, a specific wake-up word or (command) keyword of the voice control interface, etc. On this basis, the processor PRO provides a processed output OUT, which is fed to the output transducer, here the speaker SPK, to be presented to the user of the hearing device. Thus, for example, a hearing device including a voice control interface (or simply, a wake-up word detector) can be implemented. The processor PRO may also include a NN-based detector (such as an OV detector). The output of the OV detector can be an input feature of other detectors, such as a wake-up word detector. In Figure 5 In an embodiment of the invention, the output RES of the decision unit Post-Pro is fed to a processor PRO of the hearing device. Alternatively or additionally, it may be fed to another functional part of the hearing device, such as a voice interface for controlling functions of the hearing device based on the recognition of a plurality of command words. The output of the decision unit may for example be a command word or sentence or a wake-up word or sentence for activating the voice control interface. Alternatively or additionally, the output RES of the decision unit Post-Pro may be fed to a transmitter (or transceiver) to be transmitted to another device and further processed there, for example in order to activate a personal assistant (such as a smartphone or the like). The transceiver may for example receive a response from another device, such as from a personal assistant. The response may for example be used to control the hearing device, or it may be played to the user via an output transducer SPK of the hearing device.

[0168] Figure 6 An exemplary procedure is shown for subdividing a base population for providing training data for training a neural network into multiple subgroups for training multiple neural networks to provide multiple optimized neural networks, each optimized neural network representing a different characteristic of a test subject. Figure 6 It shows how the training of a neural network can be divided into different networks, each of which is trained based on a subset of the data set. Thus, multiple groups with similar speech / voice characteristics can be provided. Therefore, on its basis, a corresponding number of multiple trained networks NN1, ...NN can be generated. K .

[0169] Multiple sets of candidate networks can be generated by an iterative process. Starting with training a single NN, the people with the worst NN performance are grouped, and another NN is trained for these (and the first NN is trained for the first group of people). Alternatively, the grouping can be based on age, gender, pitch, or other similarities that measure differences between different speakers. It is thus likely that new individuals (who were not part of the training data) will perform well based on at least one trained neural network. One advantage is that the size of the neural network can be kept small, since the network does not have to be generalized to all people.

[0170] Fig. 7A An embodiment of a keyword detector according to the present invention implemented as a neural network is shown. Figure 7B shows the inclusion of Fig. 7A The input vector of the neural network is the context of the audio electrical input signal.

[0171] Fig. 7A An embodiment of a keyword spotting detector according to the present invention implemented as a neural network is shown. Fig. 7A Schematically illustrated is a deep neural network (DNN) for determining the probability p(KWq,l) of a particular keyword KWq,q=1,…,Q at a given time point (l') from an input vector of L time frames X(k,l),l=l'-(L-1),…,l' comprising an electrical input signal of a time-frequency representation (k,l) or its characteristic features (=feature vector, FV), wherein k is a frequency index and l is a time (frame) index. The electrical input signal or its characteristic features (such as cepstral coefficients or spectral features, etc.) at the current time l=l', at Figure 1 , 2 It is called the "feature vector FV" in Fig. 7A , 7B The L (last) time frames of the input signal (X(k,l)) constitute an exemplary input vector of the neural network at a given time point l=l' and are recorded in Fig. 7A , 7B In Chinese This "context" included in each input vector is Figure 7B The keyword detection detector can be configured such that only the parameters of the last q layers differ (between NN candidate networks).

[0172] The current time frame (l') and the L-1 previous time frames are stacked into a vector and used as the input layer in the neural network (collectively denoted as See also Figure 7BEach time frame X(k, l') includes K values ​​(e.g. K = 16 or K = 24 or K = 64 or K = 128) of the electrical input signal (or features extracted therefrom). The signal can be represented by its magnitude |X(k, l')| (e.g. by ignoring its phase ), see Figure 7B . Alternatively, the input vector may comprise time samples of the input signal (time domain) covering an appropriate time period. The appropriate number of time frames is related to the correlations inherent in speech. In an embodiment, the L-1 previous time frames considered together with the current time frame l=l' may, for example, correspond to a time period lasting more than 20 ms, such as more than 50 ms, such as more than 100 ms, such as about 500 ms. In an embodiment, the number of time frames considered (=L) is greater than or equal to 4, such as greater than or equal to 10, such as greater than or equal to 24, for example in the range of 10-100. In the present application, the width of the neural network is equal to K·L, which for K=64 and L=10 means that the input layer L1 consists of N L1 = 640 nodes (representing 32 ms time segments of the audio input signal (for a sampling frequency of 20 kHz and 64 samples per frame, assuming no overlap in time frames)). The number of nodes (N) in the subsequent layers (L2, ..., LN) L2 ,…,N LN ) can be greater or less than the number of nodes N in the input layer L1 L1 , and in general, adapted to the respective application (taking into account the available number of input data sets and the number of parameters to be estimated by the neural network). For applications in portable hearing devices with limited power and space, the subsequent layers (N L2 ,…,N LN ) may preferably include fewer (e.g., significantly fewer) nodes, such as nodes of the order of output nodes. In this example, the number of nodes N in the output layer LN is LNis Q (such as ≤20, or 10 or less) as it includes Q values p(KWq,l’) (q = 1, …, Q) of a probability estimator, one value for each of the Q keywords of the voice interface. Optionally, the output layer may include Q+1 or Q+2 nodes by including a value for the detection of the user's own voice and / or for the detection of a “filler” (no keyword). In an embodiment, whenever the filter bank of the hearing device provides a new time frame of the input signal, the neural network is fed a new input feature vector (i.e., in this case, there will be a certain time frame overlap between the input vectors). However, to reduce computational complexity (and power consumption), the neural network may be executed less frequently than once per time frame, for example once every 10 time frames, or less than once every 20 time frames (such as less than once every 20 ms or less than once every 40 ms). However, preferably, the context (input feature vector) fed to the neural network at a given time point overlaps (in time) with the previous context. In an embodiment, the number of time frames ΔL between each new execution of the neural network is less than the number of time frames L in the input feature vector (ΔL < L, for example ΔL / L ≤ 0.5) to ensure context overlap. As an alternative to stacking time frames, a recurrent network structure (such as an LSTM or GRU network) may be utilized. Thereby, the input layer can be significantly smaller.

[0173] Fig. 7A For illustrating any type of general multi-layer neural network, such as a deep neural network, herein embodied as a standard feed-forward neural network. The depth (number of layers) of the neural network is denoted as N in Fig. 7A and can be any number, typically adapted to the application involved (e.g., limited by the size and / or power capacity of the device involved such as a portable device e.g., a hearing aid). In an embodiment, the number of layers in the neural network is greater than or equal to 2 or 3. In an embodiment, the number of layers in the neural network is less than or equal to 10, for example in the range of 2 to 8 or in the range of 2 to 6.

[0174] Fig. 7A The nodes of the neural network illustrated in v,u are used to implement the standard function of the neural network such that the value of the branch from the previous node to the node involved is multiplied by the weight associated with the corresponding branch and the contributions are added together as the sum value Y’ of node v in layer u v,u . The sum value Y’ uv then undergoes a non-linear function f, thereby providing the composite value Z of node v in layer u v,u = f(Y’ Fig. 7A ). This value is fed to the next layer (u+1) via the branch connecting node v in layer u to the nodes in layer u+1. In v,u , the sum value Y’ of node v in layer u

[0175]

[0176] where w p,v (u) refers to the weight of node v in layer u that will be applied to the input from node p in layer L(u-1), and Z p,v (u-1) is the signal value of the pth node in layer u-1. The same activation function f is used for all nodes (but this is not necessary). The nonlinear function can be parameterized, and one or more parameters of the nonlinear function can be included in the optimization of the node parameters. In addition, the bias parameter b p,v Can be associated with each node and participate in the optimization of node parameters. An exemplary nonlinear activation function Z = f(Y) is Fig. 7A Schematically shown in the illustration in . Typical functions used in neural networks are rectified linear units (ReLu), hyperbolic tangent functions (tanh), sigmoid or softmax functions. However, other functions can also be used. As indicated, activation functions such as ReLu functions can be parameterized (e.g. to allow different slopes).

[0177] The (possibly parameterized) activation functions f of the different layers of the neural network together with the weights w and the bias parameters b constitute the parameters of the neural network. They represent the parameters that are (together) optimized in a corresponding iterative procedure of the neural network of the invention. The same activation function f can be used for all nodes (so that the "parameters of the neural network" are constituted by the weights and bias parameters of the various layers). In an embodiment, at least for some of the nodes of the neural network, no activation function f is used.

[0178] Typically, the candidate neural network according to the invention is optimized (trained) in an offline procedure, for example using a model of a human head and torso (such as from Brüel & The HATS model is a head and torso simulator (HATS) 4128C from Sound & Vibration Measurement A / S, where the HATS model is "equipped" with a hearing device (or pair of hearing devices) of the same type (style) as the hearing device the user plans to use. The hearing device is configured to pick up (acoustically transmitted) training data when positioned at the model's ear (as during normal use of the hearing device by the user). (e.g. based on Figure 6 The scheme) defines multiple different categories of testers or based on tester parameters such as age, gender, weight / height ratio, occupation, "body shape", etc., based on the involved (N x ) different training data for each group of personnel, optimize N x Ideally, training data related to the user's normal behavior and experience of the acoustic environment should be used.

[0179] In the case of training different networks based on personalized acoustic properties, it is best to record acoustic properties from different people. Individual acoustic properties can be obtained, for example, as described in [Moore et al., 2019].

[0180] For keyword detector applications, self-speech detection can be advantageously used to define where to look for keywords in the user's sentences. Thus, the self-speech detection signal can be used as a pre-processor ( Figure 1 , 2 , 5) to define the electrical input signals (IN1, IN2) from the microphones (M1, M2). As an alternative, the self-voice detection signal may form part of the feature vector used as input to the neural network. This may be advantageous because the user is unlikely to plan to trigger a keyword (such as a wake-up word or a command word) in the middle of a sentence. The use of a self-voice presence indicator may allow the keyword to be detected only at the beginning of a sentence. For example, a rule may be imposed that a keyword may only be (validly) detected if the self-voice was not detected in the previous 0.5 seconds, or the previous 1 second, or the previous 2 seconds (but is "now" detected).

[0181] exist Fig. 7A In , the neural network is illustrated as a feed-forward network, but other neural network configurations may also be used, such as a convolutional network (CNN), a recurrent network, or a combination thereof.

[0182] Figure 8 An embodiment of a hearing device according to the invention is shown, comprising an adaptation unit configured to enable selection of a set of optimization parameters for a neural network among a plurality of optimization parameters without using external equipment. Figure 8 A self-contained hearing device HD, such as a hearing aid, according to the invention is shown, comprising an optimized neural network NN*, for example for implementing a detector DET that affects processing of the hearing device (see control signal RES to the signal processor PRO). Figure 2 Same functionality as shown and described, but without the need for external equipment ( Figure 2 ExD in the neural network node parameter optimization program (thus no wireless link to external devices is required ( Figure 2 WLNK in )). The optimized candidate neural network (NN1, NN2, …, NN K ) and the evaluation unit (comparison and selection CMP-SEL) are included in the hearing device (see adaptation unit ADU). Figure 8In the embodiment of , two microphones (M1, M2) are shown, each providing an electrical input signal representing the sounds in the environment. Other numbers of input transducers such as microphones may also be used, for example one, three or more. An input transducer from an auxiliary device such as a hearing device mounted at the contralateral ear (or any other body-worn microphone) may also provide input features to the neural network. A beamformer may be included in the pre-processor Pre-PRO to enable the generation of a directional signal based on the electrical input signals (IN1, IN2) from more than two microphones. The directional signal may, for example, be or include an estimate of the user's own voice (e.g. generated by a self-voice beamformer directed toward the user's mouth). The beamformed signal (or its characteristic features) may be a signal fed to a neural network for implementing a detector such as a self-voice detector or a keyword detector (see feature vector FV). The optional sensor SENSE providing the sensor control signal SIN to the pre-processor Pre-PRO may or may not form part of the hearing device HD. The sensor may, for example, be a motion sensor, for example including an acceleration or gyroscope sensor. Other sensors may for example be or include a magnetometer, an electroencephalogram (EEG) sensor, a magnetoencephalogram (MEG) sensor, a heart rate detector, a plethysmogram (PPG) sensor, etc. The electrical input signal (IN1, IN2) is fed to a processor PRO. The processor PRO may be configured to apply a frequency and / or level dependent gain to the electrical input signal (or a processed version thereof, such as a spatially filtered (beamformed) version thereof). The processor PRO provides a processed output signal OUT which is fed to an output transducer, here a loudspeaker SPK, for presentation to a user of the hearing device.

[0183] The structural features of the apparatus described above, described in detail in the “Detailed Description of the Invention” and defined in the claims may be combined with the steps of the method of the present invention when appropriately replaced by corresponding processes.

[0184] Unless expressly stated, the singular forms "one", "the" used herein include the plural form (i.e., have the meaning of "at least one"). It should be further understood that the terms "having", "including" and / or "comprising" used in the specification indicate the presence of the described features, integers, steps, operations, elements and / or parts, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, parts and / or combinations thereof. It should be understood that, unless expressly stated, when an element is referred to as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be an intermediate intervening element. As used herein, the term "and / or" includes any and all combinations of one or more listed related items. Unless expressly stated, the steps of any method disclosed herein do not have to be performed in the exact order disclosed.

[0185] It should be appreciated that reference to "an embodiment" or "embodiment" or "aspect" or features that "may" be included in this specification means that the specific features, structures or characteristics described in conjunction with the embodiment are included in at least one embodiment of the present invention. In addition, the specific features, structures or characteristics may be appropriately combined in one or more embodiments of the present invention. The foregoing description is provided to enable those skilled in the art to implement the various aspects described herein. Various modifications will be apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects.

[0186] The claims are not limited to the various aspects shown herein, but rather have the full scope consistent with the claim language wherein, unless expressly stated otherwise, elements referred to in the singular do not mean "one and only one" but rather "one or more." Unless expressly stated otherwise, the term "some" means one or more.

[0187] Therefore, the scope of the present invention should be judged based on the following claims.

[0188] References

[0189] [Moore et al., 2019] Moore, AH, de Haan, JM, Pedersen, MS, Naylor, PA, Brookes, M., & Jensen, J. (2019). Personalized signal-independent beamforming forbinaural hearing aids. The Journal of the Acoustical Society of America, 145(5), 2971-2981.

Claims

1. A hearing device configured to be located at or in an ear of a user or to be fully or partially implanted in a user's head, the hearing device comprising: - an input transducer comprising at least one microphone for providing at least one electrical input signal representative of sounds in the environment of the hearing device; - a preprocessor, configured to process the at least one electrical input signal and provide a plurality of feature vectors, each feature vector representing a time period of the at least one electrical input signal; a neural network processor adapted to implement a neural network for implementing the detector or a portion thereof, configured to provide an output indicative of a characteristic characteristic of at least one electrical input signal, the neural network comprising an input layer, an output layer and a plurality of hidden layers, each layer comprising a plurality of nodes, each node being defined by a plurality of node parameters, the neural network being configured to receive the plurality of feature vectors as input vectors and to provide a corresponding output vector representing an output of the detector or a portion thereof based on the input vectors; - a post-processor configured to receive the output vector, and wherein the post-processor is configured to process the output vector and provide a resulting signal; - a transceiver comprising a transmitter and a receiver for establishing a communication link to another part or another device or a server, said communication link enabling the transmission of data to and the reception of data from the other part or another device or a server at least in an adaptive operating mode; - a selector for sending the feature vector to the transmitter for transmission to another part or another device or server in the adaptive operating mode, and for sending the feature vector to the neural network processor for use as input to the neural network in the normal operating mode; the hearing device being configured to receive optimized node parameters from another part or another device or a server in an adaptive operating mode and to apply the optimized node parameters to nodes of the neural network so as to implement the optimized neural network in the neural network processor, wherein the optimized node parameters have been selected among a plurality of sets of node parameters for a corresponding plurality of candidate neural networks according to a predetermined criterion based on the feature vector; Wherein, the predetermined criterion is based on a prompt word or sentence spoken by a user, and the optimized node parameters are selected by comparing output vectors from multiple candidate neural networks with corresponding node parameters based on the prompt word or sentence spoken by the user.

2. A hearing device according to claim 1, comprising a sensor for sensing a property of a user or an environment of the hearing device and providing a sensor signal representing a current value of the property of the environment, wherein the sensor signal is an input to the pre-processor. 3 . The hearing device according to claim 2 , wherein the pre-processor is configured to process the at least one electrical input signal and the sensor signal to provide a feature vector.

4. The hearing device of claim 1, comprising an output transducer for presenting the processed output signal to a user as a stimulus perceivable as sound.

5. The hearing device according to claim 1, comprising an analysis filter bank for converting a time domain input signal into a plurality of sub-band signals to provide an input signal in a time-frequency representation (k, l), wherein k and l are frequency and time indices, respectively. 6 . The hearing device according to claim 1 , wherein the pre-processor is configured to extract features of at least one of the electrical input signal and / or the sensor signal.

7. A hearing device according to claim 2, wherein the time period of the at least one electrical input signal and optionally the corresponding value of the sensor signal covered by a given feature vector is used as input to an input layer of a neural network, which comprises at least one time frame of the at least one electrical input signal.

8. The hearing device according to claim 1, wherein the detector implemented by the neural network or a part thereof is or comprises a own voice detector and / or a keyword detector.

9. The hearing device according to claim 1, consisting of or comprising a hearing aid, a headset, an earphone, an ear protector or a combination thereof.

10. The hearing device according to claim 1, wherein the predetermined criterion is related to minimizing a cost function with respect to the output vector.

11. The hearing device according to claim 1, wherein: The predetermined criterion is based on the performance of the neural network in terms of true positives, false positives, true rejections and false rejections of the output vectors when the plurality of feature vectors are extracted from a time segment of at least one electrical input signal having known characteristics.

12. A method for selecting optimized parameters of a neural network for use in a portable hearing device, the method comprising: - providing a portable hearing device to be used by a specific user, the hearing device comprising a neural network processor adapted to implement a neural network comprising an input layer, an output layer and a plurality of hidden layers, each layer comprising a plurality of nodes, each node being defined by a plurality of node parameters and a non-linear function, the neural network being configured to receive an input vector and to provide an output vector which is a specific non-linear function of the input vector; - fitting the hearing device to the user; - providing at least one electrical input signal representative of sounds in the environment of a user wearing the hearing device; - processing the at least one electrical input signal and providing a plurality of feature vectors, each feature vector representing a time period of the at least one electrical input signal; - providing a plurality of pre-trained candidate neural networks, wherein each pre-trained network is considered as a candidate network of the user, and wherein each pre-trained neural network has been trained based on completely or partially different training data; - receiving the feature vector and providing it as an input vector to the plurality of pre-trained candidate neural networks; - determining corresponding output vectors corresponding to the feature vectors from the plurality of pre-trained candidate neural networks; - comparing the output vectors and selecting one of a plurality of candidate neural networks as a neural network optimized for the hearing device according to a predetermined criterion on the output vectors, wherein the predetermined criterion is based on a cue word or sentence spoken by the user, the optimized neural network being selected by comparing the output vectors from a plurality of candidate neural networks having corresponding node parameters based on the cue word or sentence spoken by the user; - transmitting node parameters of a selected neural network from a plurality of candidate neural networks to a neural network processor of the hearing device, and - receiving the node parameters into the neural network processor and applying them to the neural network.

13. The method of claim 12, wherein each candidate neural network has been trained based on training data from different classes of people exhibiting different acoustic properties.

14. The method according to claim 12, wherein a signal representing a current value of a characteristic of the user or the user's environment is provided in the hearing device together with at least one electrical input signal and processed to provide the feature vector.

Images