Hearing device fitting agent with customized environment model

Abstract

A fitting agent for a hearing device system is disclosed, the hearing device system comprising a hearing device worn by a hearing device user, where the fitting agent comprises one or more processors configured to: initialize a user model; initializing an environment model; obtaining environmental data indicating a current environment; determining an environmental state based on the environmental data using the environmental model; obtaining test settings for the hearing device based on the environmental state, the test settings including a test setting A and a test setting B; presenting the test settings to the hearing device user; obtaining a user input of a preferred test setting indicative of a preference for either the A test setting or the B test setting; updating the user model based on the preferred test settings and the environmental state to provide an updated user model; and updating the environmental model based on the preferred test settings and one of the user model and the updated user model to provide an updated environmental model.

Description

Technical Field

[0001] The present disclosure relates to hearing devices and related tools, methods and systems, in particular tools, methods and systems for determining, tuning, fitting and optimizing one or more of hearing device parameters and models thereof. Accordingly, a fitting agent for a hearing device system including a hearing device is provided. Background Art

[0002] Fitting and tuning of hearing devices or hearing aids has always been considered a tedious task for healthcare professionals (HCPs). Traditional methods for fitting hearing device parameters rely on compensating for the user's hearing loss by applying rules such as NAL-NL1 or NAL-NL2 based on audiograms. However, these rules do not take into account specific user preferences.

[0003] Recent approaches involve preference learning for hearing devices. Summary of the invention

[0004] Challenges remain in improving tools, methods and equipment to allow for improved fitting and tuning of hearing device parameters.

[0005] A fitting agent is disclosed, which is optionally used in a hearing device system including a hearing device worn by a hearing device user. The fitting agent includes one or more processors, which are configured to: initialize a user model, the user model including multiple user preference functions and associated user response distributions; initialize an environment model; obtain environment data indicating a current environment; use the environment model to determine an environment state based on the environment data; based on the environment state, obtain test settings for the hearing device, the test settings including test setting A and test setting B; present the test settings to the hearing device user; obtain user input of a preferred test setting indicating a preference for test setting A or test setting B; based on the preferred test setting and the environment state, optionally update the user model to provide an updated user model; and based on the preferred test setting and optionally based on one or more of the user model and the updated user model, update the environment model to provide an updated environment model.

[0006] The present disclosure takes into account any differences in user preferences in different environments and provides efficient and memory-saving user preference model updates. Specifically, there is no need to store historical specific user feedback (previous A test setting and B test setting and their preferences), thereby reducing memory requirements.

[0007] Furthermore, the context-dependent user preference functionality may provide a simpler user preference functionality, thereby allowing a simple and efficient way to configure one or more hearing device parameters of a hearing device. Furthermore, by improving the modeling of user-preferred hearing parameter settings in different environments, thereby enabling the application of optimized settings in the hearing device, the present disclosure provides an improved hearing experience for the user, which in turn allows an improved user experience.

[0008] Furthermore, by incorporating user feedback into the learning cycle, the present disclosure provides an efficient automatic search for optimal hearing device parameters. A fitting agent is provided that allows learning user preferences for hearing device parameters in an efficient and minimally intrusive manner by empowering the user to make decisions directly and have a direct impact on the fitting and / or tuning process.

[0009] Furthermore, the present disclosure allows for configuration (eg, fitting and / or tuning) of hearing device parameters during normal operation and / or with little user input / interaction. Thus, a simple and smooth user experience of the hearing device is provided.

[0010] An advantage of the present disclosure is that a hearing device user receives fully personalized settings for his or her hearing device based on the user's own preferences in a personally customized environment. The user's preferences are used to define and adjust the user's environment and therefore also the corresponding hearing device settings or operating modes, which in turn improves user satisfaction and leads to more frequent use of the hearing device.

[0011] The present disclosure provides environmental personalization / customization with improved system performance. This is achieved by introducing a bidirectional dependency between the environmental domain and the (user) preference domain, thereby providing significant performance improvements over a baseline model (i.e., a fitting agent without user feedback into the environmental model). BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The above and other features and advantages of the present invention will be readily understood by those skilled in the art through the following detailed description of exemplary embodiments of the present invention with reference to the accompanying drawings, in which:

[0013] Figure 1 schematically illustrates a hearing system according to the present disclosure,

[0014] Figure 2 shows the interaction between the user and the fitting agent, and

[0015] Figure 3 is a flow chart of an exemplary method according to the present disclosure. DETAILED DESCRIPTION

[0016] Various exemplary embodiments and details are described below with reference to the accompanying drawings when relevant. It should be noted that the drawings may or may not be drawn to scale, and the same reference numerals are used throughout the drawings to represent elements of the same structure or function. It should also be noted that the drawings are only for the convenience of describing the embodiments. The drawings are not intended as an exhaustive description of the present invention, or as a limitation of the scope of the present invention. In addition, the illustrated embodiments do not need to have all the aspects or advantages shown. The aspects or advantages described in conjunction with a particular embodiment are not necessarily limited to that embodiment and may be implemented in any other embodiment even if not shown as such or not explicitly described.

[0017] A fitting agent is disclosed. The fitting agent or at least a first part thereof can be implemented in an accessory device (e.g., an electronic device), for example as an application. The accessory device includes an interface, a processor, and a memory. For example, the accessory device can be or include a mobile phone (e.g., a smartphone), a smart watch, a dedicated device, a computer (e.g., a laptop or PC), or a tablet computer. The fitting agent or at least a second part thereof can be implemented in a server device. The fitting agent or at least a third part thereof can be implemented in a hearing device. In other words, the fitting agent can be implemented in one or more electronic devices (e.g., a hearing device and / or an accessory device).

[0018] The present disclosure relates to a hearing device system, a fitting agent for a hearing device system, an accessory device and a hearing device and related methods. The accessory device forms an accessory device of a hearing device. The accessory device is typically paired or wirelessly coupled with a hearing device. The hearing device may be, for example, a behind-the-ear (BTE), in-the-ear (ITE), in-the-canal (ITC), receiver-in-the-canal (RIC), receiver-in-the-ear (RITE) or microphone-in-the-ear (MaRIE) hearing aid. The hearing device may be a hearable device, such as a pair of earplugs or a headset. Typically, the hearing device system is owned and controlled by a hearing device user.

[0019] The hearing device system may include a server device and / or a fitting device. The fitting device is controlled by a dispenser and may be configured to determine configuration data (eg fitting parameters). The server device may be controlled by a hearing device manufacturer.

[0020] The fitting agent may be a fitting agent for a hearing device system comprising a hearing device worn by a hearing device user. The fitting agent may also be referred to as a hearing device fitting agent.

[0021] The fitting agent includes one or more processors. The one or more processors are configured to: initialize a user model and an environment model. The environment model is also referred to as a model of the environment. The user model and / or the environment model can be obtained from a memory of the fitting agent.

[0022] The user model comprises a plurality of user preference functions and associated user response distributions. The user response distributions are optionally state-dependent and can be expressed as a single distribution where, depending on the state, the distribution has a corresponding state-dependent form.

[0023] The environment model defines or includes (at least after a step of user feedback) a plurality of active (observed) environment categories or clusters, and / or inactive environment categories or clusters. Each active environment category may correspond to a (single) user preference function. The inactive environment categories are optionally infinite and may be bundled in a single inactive category or cluster. Thus, new clusters may always be spun-out from an inactive cluster as active clusters. Thus, it may be said that the environment model includes an infinite number of categories or clusters. It should be noted that categories and clusters are used interchangeably herein.

[0024] The user model optionally represents one or more probabilistic descriptions of user responses when comparing two sets of hearing device parameter settings. The main part of the user model includes one or more (e.g., plural) user preference functions and one or more (e.g., plural) user responses to the presented parameter selections (also referred to as user response distributions). The user model includes one or more distributions of parameters of the corresponding user preference functions. In other words, each user preference function has an associated distribution of hearing device parameters. In other words, the user model may include a first user preference function and an associated first user response distribution, wherein the first user preference function is optionally associated with a first environment category or cluster of the environment model. The user model may include a second user preference function and an associated second user response distribution, wherein the second user preference function is optionally associated with a second environment category or cluster of the environment model. The user model may include a third user preference function and an associated third user response distribution, wherein the third user preference function is optionally associated with a third environment category or cluster of the environment model.

[0025] A vector of hearing device parameters is optionally defined on a D-dimensional continuous compact surface. Specifically, a hearing device parameter x is optionally defined on a D-dimensional hypercube, i.e., x∈[0,1] D In one or more exemplary fitting agents, the hearing device parameters may be normalized to their physical ranges. The fitting agent is configured to find optimized / improved values ​​of hearing device parameters (also denoted as θ) for a particular user. The number D of hearing device parameters may be 1 and / or less than 100, for example in the range of 10 to 50. The number D of hearing device parameters may be greater than 20, for example in the range of 25 to 75.

[0026] Usually, the user preference function is unknown. User preference function (f(x,θ,Λ,k) = f(x;θ k ,Λk ), k = 1, 2, ..., K, where K is the number of user preference functions) can be hearing device parameters x∈[0,1] D , which has a known form but unknown shape. This shape can be optionally obtained by fitting or tuning the parameter θ k ∈[0,1] D and the scaling matrix Λ k The scaling matrix Λ k It can be a positive definite scaling matrix Λ k The scaling matrix can be a diagonal matrix

[0027] The user preference function or at least one or more of the K user preference functions may be unimodal.

[0028] In one or more exemplary fitting agents, a distribution prior may be applied to Λ k For example, we can k A lognormal prior is applied to each element of , e.g., Or you can k Apply a Gamma distribution prior to each element of d ~Gamma(α d ,β d ).

[0029] Scaling matrix Λ k It does not have to be a diagonal matrix. The scaling matrix Λ k Select Λ k = L′*L, where L is a lower triangular matrix (also called Λ k Cholesky decomposition of ). We can apply a Gaussian prior to each element of L, for example,

[0030] The maximum parameter θ of the kth environment k can be a transformation of another variable, for example it can be constrained by the following a priori assumptions:

[0031]

[0032] in, is the cumulative density function of a probability distribution (e.g., the standard normal distribution), and is a sample from another probability distribution.

[0033] In one or more example fitting agents / methods, maximizing the parameter θ k It can be constrained by the following assumptions: k Expressed as Transformation of and the corresponding prior distribution of the transformed variable:

[0034]

[0035] in, is the cumulative density function of the standard normal distribution, is from a vector with mean μ k and the covariance matrix Σ k The mean and covariance values ​​can be learned from the user responses.

[0036] In one or more example fitting agents / methods, the scaling matrix may be a transformed scaling matrix, such as by applying a logarithmic transformation to the elements of the scaling matrix.

[0037] The user preference function can be expressed as f or f k (x;θ k ,Λ k ), where each user preference function f k (x;θ k ,Λ k ) is associated with an environment or state k of K environment / user preference functions. User preference function f k (x;θ k ,Λ k ) can be given by the following formula:

[0038] f k (x;θ k ,Λ k )=-((x-θ k ) T Λ k (x-θ k )) v (C)

[0039] where x is optionally in the hypercube [0,1] D is a D-dimensional vector in which the (D) hearing device parameters of the device in the kth environment are represented; θ k Yes k The maximization independent variable of k is a positive definite D×D scaling matrix characterizing the user sensitivity to changes in hearing device parameters; D is an integer; and v is a real-valued exponent. The real-valued exponent v may be in the range of 0.01 to 0.99. The real-valued exponent v may be less than 0.5, for example in the range of 0.01 to 0.45. The real-valued exponent v may be greater than 0.5, for example in the range of 0.55 to 0.99. In one or more exemplary fitting agents / methods, the real-valued exponent v may be in the range of 0.25 to 0.75, for example 0.5.

[0040] User preference function f k (x;θ k ,Λ k ) can be given by the following formula:

[0041] f(x;θ k ,Λ k )=exp(-(x-θ k ) T Λ k (x-θ k )).

[0042] Variable θ k ,Λ k Also known as the user preference function parameter, where θ k is the optimal value of the user preference function corresponding to the optimal user-specific hearing device parameters, and is a diagonal matrix indicating the user sensitivity (ie, the ability to discern parameter changes). It should be noted that other unimodal user preference functions are also possible.

[0043] The fitting agent is configured to: obtain a test setting (also denoted as x_ref, x ref or θ t ). A test setting x_ref is a vector comprising D hearing device parameters for a hearing device. The hearing device parameters may include one or more of the following: filter coefficients, compressor settings, gains or other parameters related to the operation of the hearing device or signal processing in the hearing device. The A test setting may be based on and / or dependent on the current environment as indicated by the environmental state determined via the environmental model. In other words, the A test setting may be based on and / or dependent on the environmental state, also denoted c(t), t indicating the time step.

[0044] The fitting agent is configured to: obtain a test setting (also denoted as x_alt, x alt or θ' t ). The B test setting x_ref is a vector comprising D hearing device parameters for the hearing device. The hearing device parameters may include one or more of the following: filter coefficients, compressor settings, gains or other parameters related to the operation of the hearing device or signal processing in the hearing device. The B test setting may be based on and / or dependent on the current environment as indicated by the environmental state determined via the environmental model. In other words, the B test setting may be based on and / or dependent on the environmental state.

[0045] The fitting agent (e.g., one or more processors of the fitting agent) is configured to present the test settings to the hearing device user. Presenting the test settings to the hearing device user includes: presenting the A test setting and / or the B test setting to the user, for example, by sending the A test setting and / or the B test setting to the hearing device; the hearing device processes the incoming sound according to the A test setting and / or the B test setting so as to present the A test setting and / or the B test setting to the user, thereby allowing the user to evaluate the test settings. Presenting the A test setting and the B test setting to the user optionally includes: outputting the A test signal (also represented as y) according to the A test setting. t ). Test signal A may be an audio signal. Test signal A may be output via a speaker or a receiver of the hearing device. Presenting test setting A and test setting B to the user may optionally include: generating test signal A in an accessory device according to test setting A, and streaming test signal A from the accessory device to the hearing device. Presenting test setting A and test setting B to the user may optionally include: sending a control signal indicating test signal A / test setting A from the accessory device to the hearing device. The control signal may include test setting A. Presenting test setting A and test setting B to the user may include: generating test signal A in the hearing device according to the control signal (e.g., test setting A based on the control signal).

[0046] Presenting the A test setting and the B test setting to the user may optionally include: outputting the B test signal (also represented by y') according to the B test setting. t ). Test signal B may be an audio signal. Test signal B may be output via a speaker or a receiver of the hearing device. Presenting test setting A and test setting B to the user may optionally include: generating test signal B in an accessory device according to test setting B, and streaming test signal B from the accessory device to the hearing device. Presenting test setting A and test setting B to the user may optionally include: sending a control signal indicating test signal B / test setting B from the accessory device to the hearing device. The control signal may include test setting B. Presenting test setting A and test setting B to the user may include: generating test signal B in the hearing device according to the control signal (e.g., test setting B based on the control signal).

[0047] The fitting agent (eg, one or more processors of the fitting agent) is configured to obtain (eg, receive and / or detect) user input of a preferred test setting indicating a preference for the A test setting or the B test setting.

[0048] A non-preferred test setting is one in which test setting A or test setting B is not selected as a preferred test setting. Therefore, a hearing device and / or an accessory device that implements a fitting agent or at least a part of a fitting agent may include one or more user interfaces for obtaining (e.g., receiving and / or detecting) user input. For example, a hearing device may include a user interface for receiving user input. The user interface of the hearing device may include one or more buttons, an accelerometer, and / or a voice control unit. The accessory device may include a user interface. The user interface of the accessory device may include a touch-sensitive surface (e.g., a touch display) and / or one or more buttons. The user interface of the accessory device may include a voice control unit. The user interface of the hearing device may include one or more physical sliders, knobs, and / or push buttons. The user interface of the accessory device may include one or more physical or virtual (on-screen) sliders, knobs, and / or push buttons.

[0049] The fitting agent is configured to obtain (e.g., detect) a user input of a preferred / selected test setting indicating a preference for test setting A or test setting B. In the fitting agent, detecting the user input of the preferred test setting indicating a preference for test setting A or test setting B may include prompting a user for the user input, e.g., via a beep signal or a voice signal from the hearing device, and / or a visual, tactile, and / or audio prompt from the accessory device. Detecting the user input may be performed at the hearing device, e.g., by the user activating a button and / or an accelerometer in the hearing device (e.g., by single-clicking or double-clicking the hearing device housing). Detecting the user input may be performed at the accessory device, e.g., by the user selecting a user interface element representing the preferred test setting, e.g., on a touch-sensitive display of the user's accessory device.

[0050] In one or more examples, the fitting agent is configured to determine and update hearing device parameters of the hearing device based on the updated user model and, optionally, based on the updated environment model.

[0051] In one or more examples, a fitting agent for a hearing device system is disclosed, the hearing device system including a hearing device worn by a hearing device user, wherein the fitting agent includes one or more processors configured to: initialize a user model, the user model including multiple user preference functions and associated user response distributions; initialize an environment model; obtain environment data indicating a current environment; determine an environment state based on the environment data using the environment model; obtain test settings for the hearing device based on the environment state, the test settings including test setting A and test setting B; present the test settings to the hearing device user; obtain user input of a preferred test setting indicating a preference for test setting A or test setting B; update the user model based on the preferred test setting and the environment state to provide an updated user model; and update the environment model based on the preferred test setting and one of the user model and the updated user model to provide an updated environment model.

[0052] In one or more examples, updating the environment model includes: determining an updated environment state based on one or more of the preferred test settings, the user model, and the updated user model; and updating the environment model based on the updated environment state. In an example, the updated environment state is based on the preferred test settings and the updated user model. Therefore, updating the environment state is optionally based on the updated environment model.

[0053] In one or more examples, the updated environment state (also denoted as c'(t)) indicates a hard decision or a soft decision. In other words, the updated environment state can be an identifier of the updated environment state (hard decision) or indicate an updated environment state distribution (soft decision).

[0054] Updating the environment model may include adjusting the number of activities or observation clusters in the environment model.

[0055] In one or more examples, updating the environment model includes reducing the number of environment categories of the environment model. Reducing the number of environment categories or clusters of the environment model can be accomplished by merging environment models of at least two categories. For example, two activity clusters can be merged to form an activity cluster. For example, an activity cluster can be discarded by merging an activity cluster with an inactive cluster.

[0056] In one or more examples, updating the environment model includes increasing the number of environment categories of the environment model. Increasing the number of environment categories or clusters of the environment model can be accomplished by splitting or separating (split) one or more active categories of the environment model, for example, splitting an active environment category into two or more active environment categories, or splitting an active environment category from an inactive environment category.

[0057] In one or more examples, the environment model is an infinite Gaussian mixture model (GMM).

[0058] In one or more examples, initializing the environment model includes obtaining (e.g., acquiring from a database) a user profile, and initializing the environment model based on the user profile (e.g., by selecting an environment model from a library based on the user profile). The user profile may include one or more of age, gender, hearing loss, and activity level, and the initialization environment model is optionally based on one or more of age, gender, hearing loss, and activity level.

[0059] In one or more examples, the environment model is a Hierarchical Dirichlet Process Hidden Markov Model (HDP-HMM).

[0060] In one or more examples, the environmental categories are modeled as latent random variables, for example, from a countably infinite category space.

[0061] In one or more examples, initializing and / or updating the environment model includes modeling the environment categories using a Dirichlet process, and wherein each environment category generates a state-specific infinite Dirichlet Gaussian mixture.

[0062] In one or more examples, updating the environment model includes applying a hidden Markov model (eg, a viscous hierarchical Dirichlet process hidden Markov model).

[0063] The fitting agent / the one or more processors of the fitting agent are configured to obtain environmental data indicative of a current environment and determine an environmental state based on the environmental data, for example using an environmental model.

[0064] In one or more example fitting agents, obtaining environmental data includes obtaining audio data, and optionally determining environmental data based on the audio data and / or including the audio data in the environmental data. In other words, one or more processors of the fitting agent may be configured to obtain audio data, and determine environmental data or at least one or more environmental parameters based on the audio data. The audio data may include first audio data, which represents or indicates audio recorded by one or more microphones of a hearing device of a user. The audio data may include second audio data, which represents or indicates audio recorded by one or more microphones of one or more accessory devices of the user. The audio data may include third audio data, which represents or indicates audio wirelessly sent to the hearing device of the user. For example, the fitting agent may be configured to classify the environment based on the hearing device audio, and set one or more environmental identifiers and / or environmental probabilities of the environmental data accordingly.

[0065] In one or more example fitting agents, obtaining environmental data includes: obtaining contextual data, and optionally, determining the environmental data based on the contextual data, and / or including the contextual data in the environmental data. In other words, one or more processors of the fitting agent can be configured to: obtain contextual data, and optionally, determine the environmental data based on the contextual data. The contextual data can indicate the context in which the user is located, for example, indicating the user's location, position, movement, temperature, pulse, or other data related to the environment. For example, the contextual data may include location data (e.g., GPS coordinates) and / or movement data (e.g., accelerometer data). The contextual data may include calendar data, and the environmental data may be based on the calendar data. The contextual data may include sensor data, such as one or more sensors from an accessory device and / or one or more sensors from a hearing device. The contextual data may include hearing device data sent from a hearing device, such as one or more program identifiers, one or more operating parameters, and / or one or more operating mode identifiers of the hearing device.

[0066] In one or more example fitting agents, obtaining environment data includes: receiving user input indicating an environment, for example, via a user interface of the accessory device. Thus, a user can select and indicate a current environment via the accessory device, for example, from a list of environments presented on a touch display screen of the accessory device.

[0067] The fitting agent / one or more processors of the fitting agent are configured to update the user model, such as one or more user preference functions f based on the hearing device parameters of the preferred test settings and environmental conditions. k (x;θ k ,Λ k ) and / or one or more user response distributions P u, In other words, the user response distribution P u, User responses may be modeled for one or more environments. The user response distribution may be a weighted user response distribution for all or a subset of environments (e.g., where the weights are based on the environment probabilities ENVP_k, k=1, ..., K t (K t is the number of activity environment categories at time t)) and the user preference function f k (x;θ k ,Λ k ) parameter distribution.

[0068] The fitting agent is configured to update the user model based on the hearing device parameters of the preferred test setting, the non-preferred test setting, and optionally, the environmental state c(t). In one or more examples, the environmental state c(t) indicates a hard decision or a soft decision. In other words, the environmental state can be an identifier of a state (hard decision), or indicate an environmental state distribution (soft decision). Specifically, the fitting agent can be configured to update the user model based on (r, x_ref, x_alt, c(t)). Updating the user model may include updating the user preference function or at least its parameters, and / or updating the parameter distribution associated with the user preference function based on the environmental state and / or its reliability (soft) and one or more of the user inputs of test setting A, test setting B, and the preferred test setting. In other words, the parameters of the user preference function and / or the associated parameter distribution can be updated based on the test results including the following items: test setting A, test setting B, the preferred test setting among test setting A and test setting B, and the environmental state. Updating the user model may include: updating the user response distribution or at least its parameters based on the environment state and one or more of the user input of the test setting A, the test setting B, and the preferred test setting. In other words, the parameters of the user preference function may be updated based on the test results including the following items: the test setting A, the test setting B, the preferred test setting among the test setting A and the test setting B, and the environment data. Therefore, updating the user model may include: updating the user preference function parameters θ based on the environment state and one or more of the user input of the test setting A, the test setting B, and the preferred test setting. k and / or k Parameter distribution of .

[0069] In one or more exemplary fitting agents and / or methods, updating the user model may be based on Bayesian reasoning. Updating the user model may include updating parameters of a user preference function and / or a user response distribution and / or a parameter distribution associated with the user preference function. Updating the user model may include determining one or more posteriori for the parameters of the user preference function.

[0070] Updating the user model may include determining a posteriori for one or more (e.g., a subset) or all of the parameters in the user preference function, for example based on environmental states, previous parameter posteriori (e.g., last or current parameter posteriori), preferred test settings, and non-preferred test settings.

[0071] In one or more examples, the fitting agent is implemented as part of the preference learning system described below.

[0072] The preference learning system can be modeled as two subsystems: agent and user, which interact with each other in a given context. The goal is to learn user preferences with a minimum amount of interaction. The agent represents the trial generation process given an observed environment. The user represents the internal agent model of the actual user with his / her parameter preferences for different environments.

[0073] The user model is optionally characterized by a latent unimodal preference function parameterized by the environment state k:

[0074]

[0075] where x∈[0,1] D represents the tuning parameter; and θ k ∈[0,1] D and denote the optimal tuning parameters corresponding to the user preference and the user sensitivity to parameter changes in state k, respectively. Here, θ, Λ are the sets of all parameters corresponding to all states. It should be noted that x∈[0,1] D The assumption of can always be fulfilled by normalization.

[0076] The user interaction with the agent is defined by the user evaluation of a trial (proposed by the fitting agent). A trial consists of two parameter choices {θref, θalt} called the baseline suggestion (test setting A) and the alternative suggestion (test setting B), respectively. The user response r to a trial is defined by a binary random variable where r=1 indicating a preference for θalt and r=0 otherwise. The response is based on the unknown user preferences.

[0077] The user response model that relates binary user responses to continuous-valued tuning parameters and takes into account the user's uncertainty about their preferences is called a binomial-probit regression model. We extend this model to a state-dependent user response model:

[0078] P(r|θ ref ,θ alt ,θ,Λ,c=k)

[0079] = Ber(r|Φ(f k (θ alt ;θ,Λ)-f k (θ ref ;θ,Λ))), (2)

[0080] where c∈Ν is the state of the environment, {θref,θalt} is the trial for this state, and Ber(·|·) and Φ(·) denote the Bernoulli probability mass function (PMF) and the standard normal cumulative distribution function (CDF), respectively.

[0081] The parameters θ and Λ (all θ k and Λ k The set of parameters is the unknown latent random variables that must be learned through the interaction between the user and the agent. To operate in the unconstrained parameter domain, the parameters are transformed and redefined as φ = [Φ -1 (θ),lnΛ] T , where φ is the φ of all states k k The parameters are assumed to be Gaussian distributions a priori:

[0082] To infer these parameters, the agent can collect the most informative observations by generating trials, the responses to which yield the largest reduction in model uncertainty.

[0083] The context-dependent preference learning system is then defined by two components: one component corresponds to the generative trial distribution given previous observations and trials The other component corresponds to the predicted user response distribution used by the agent to model the user response Right now:

[0084]

[0085] It should be noted that the fitting agent / preference learning system depends on the state of the environment c t . Assume that each environment state k corresponds to a single preference function. Environment state c t (e.g., environment identifiers or environment state distributions) are determined via an environment model and fed back to a user model / preference learning system, e.g., for use in experimental design.

[0086] To model the environment, an infinite Bayesian Gaussian mixture model (GMM) approach is used, namely

[0087]

[0088] Among them, m k and W k are the mean vector and precision matrix of the kth mixture component, respectively, and their priors are assumed to be Normal-Wishart distributions with hyperparameters m0, λ0, W0, ν0:

[0089]

[0090] In addition, for the mixture probability vector π, the component or cluster label c t Follows categorical distribution

[0091] p(c t |π)=Cat(c t |π). (6)

[0092] Finally, a prior on π must be defined. It should be noted that the number of mixture components in GMM has not been defined so far. In fact, this number is unknown a priori and changes over time during the evolution of the preference learning system in user-agent interactions. The mixture components can be generated by DP, i.e., assuming K = ∞ components, where the mixture prior is given by the GEM distribution:

[0093]

[0094] In summary, the environment model is combined with the state-dependent preference learning system (user model) to obtain a context-aware preference learning system defined by the following generative model:

[0095] π~GEM(α) (8a)

[0096] c t |π~Cat(π) (8b)

[0097]

[0098] In an exemplary implementation, parameters are sequentially inferred and passed through the system from the environment model to the user preference model; and, after the trial is over, the current model state is re-evaluated and the model is updated, but now in the opposite direction: from the user preference domain (user model) to the environment domain (environment model). A sequential update and greedy search (SUGS) algorithm can be used for the DP hybrid of the environment clustering step. Cluster labels or activity environment categories can optionally be greedily selected by maximizing the cluster posterior.

[0099] The model specified by (8a)-(8d) is used to transform the initial cluster labels or environment states Estimated

[0100]

[0101] Among them, q t (c t ) is the approximate posterior, which is given by:

[0102] q t (c t )∝p(s t |ct ) t (c t ), (10)

[0103] Among them, p t (c t ) represents the relative occurrence prior of each environment:

[0104]

[0105] in, is the number of data points assigned to cluster or environment state k, and K t-1 is the total number of explored environment states up to time t-1. The evidence term can be calculated as

[0106]

[0107] Here, given the chosen conjugate prior (5), the Bayesian rule is applied. Here, b is s t Dimension.

[0108] Next, the user preference model prior is specified by (8e)-(8g). Optionally, an assumed density filter can be used to obtain an approximate preference posterior:

[0109]

[0110] For the cluster at time t represents the cluster assignment up to time t. To do this, relevant data must be collected. Therefore, an efficient experimental design strategy can be specified. First, we present a straightforward proxy strategy that can be used for each individual environment, and then propose its variant that includes cluster reliability information. The experimental design starts with an initial experiment, for example, randomly selected A surrogate strategy for efficient sequential experimental design (aiming to rapidly reduce model uncertainty) is defined as

[0111]

[0112] Where I(·; ·) represents mutual information, R t Respond to RV, is the test result at time t, and is the marginalized posterior of the sensitivity parameter (13).

[0113] The above experimental design is based on difficult decisions about cluster assignments. In this example, it is proposed to incorporate soft information about cluster assignments into the design of replacement suggestions. The replacement suggestion is given by:

[0114]

[0115] According to (14), k = 1, 2, ..., K t generate and is generated by new unexplored agents and is therefore sampled from the preference prior. After experimentation, given new data, we need to update our model.

[0116] It should be noted that a replacement experimental design can be applied by maximizing weighted mutual information based on the probability distribution of the environment clusters.

[0117] By evaluating the new benchmark The posterior membership of (also called the personalized environment state) is determined by the user response, allowing the cluster assignment decision (the estimate of the environment state) to be re-evaluated. The updated personalized environment state is then given by:

[0118]

[0119] in, The approximate posterior distribution of is given by a preference mixture:

[0120]

[0121] here, (as before) is the marginalized parameter posterior, and is the prior on the relative occurrence of each personalized cluster; or the prior as the posterior of the environment state from the environment model. Finally, the new environment state is used as cluster labels for environmental data (e.g., feature signals s t ), and is used to use s t Update the environment model and use D t renew Alternatively, to incorporate soft (reliability) information, we can use q t (.) to update the user model and / or environment model.

[0122] Figure 1 A schematic overview of a fitting agent with a context-aware preference learning system is shown. A signal s is received as environmental data by the hearing device 2 and in a fitting agent 27 (eg implemented in an accessory device). t , the fitting agent passes the environment data to the environment model 50. The environment model 50 is based on the audio input signal s t Environmental data to determine the environmental status c t(e.g., an environment identifier and / or environment distribution associated with the environment 41), and passes the environment state to the preference learning system 52. Based on the environment state and previous trials and user responses, the preference learning system 52, which includes a user model 54, proposes two tuning parameter sets θ (test setting A) and θ′ (test setting B) for the hearing device. The hearing device 2 uses these tuning parameters θ and θ′ to process the signal s observed by the microphone of the hearing device 2. t , to generate audio signals y and y', respectively, which are then presented to the user 40 via the receiver of the hearing device for use in evaluating the test settings. The user 40 indicates a preference for the A test setting or the B test setting via a user input represented by r that is fed back to the preference learning system 52. The preference learning system 52 updates the user model 54 and updates the updated environment state c' t (eg, an updated environment identifier associated with the environment 41 and / or an updated environment distribution) is fed back to the environment model to update the environment model for providing an updated environment model 50' based on the preferred test settings and one of the user model or updated user model.

[0123] Figure 2 is an overview of a hearing system with a fitting agent according to the present disclosure. The hearing system comprises a hearing device 2, an accessory device 4 and an optional server device / fitting device 5. The hearing device 2 comprises a transceiver module 6 for communicating with the accessory device 4 and an optional contralateral hearing device ( Figure 1 The transceiver module 6 comprises an antenna 8 and a transceiver 10, and is configured to receive and / or send wireless signals with the accessory device 4 via a wireless connection 11. The accessory device 4 is configured to receive and / or send wireless signals with the server device / fitting device 5 via a wireless connection 11A. The hearing device 2 comprises: a set of one or more microphones, including a first microphone 12 for providing a first microphone input signal 14; a processor 16 for processing an input signal including the first microphone input signal 14 according to one or more hearing device parameters, and providing an electrical output signal 18 based on the input signal; an optional user interface 20 connected to the processor 16; and a receiver 22 for converting the electrical output signal 18 into an audio output signal.

[0124] The accessory device 4 is a smartphone and includes a user interface 24 including a touch display screen 26, a processor (not shown) and a memory (not shown).

[0125] In the hearing device system 1 , the fitting agent 27 is an application installed in the memory of the accessory device 4 .

[0126] The fitting agent 27 is a fitting agent for updating a user model and / or an environment model of a hearing device user and / or for one or more of optimization, determination, fitting, tuning and modeling of hearing device parameters of a hearing device.

[0127] The fitting agent 27 includes one or more processors (e.g., one or more processors of the accessory device and the hearing device), which are configured to: initialize a user model and an environment model, the user model including a plurality of user preference functions and associated user response distributions, wherein each user preference function f(x; θ k ,Λ k ) and the kth environment (k=1,2,...,K t ) association; initializing an environment model defining a plurality of environment categories; obtaining environment data indicating a current environment; using the environment model to determine an environment state based on the environment data; obtaining a test setting for a hearing device based on the environment state, the test setting comprising a test setting A and a test setting B; presenting the test setting to a hearing device user; obtaining user input of a preferred test setting indicating a preference for the test setting A or the test setting B; updating a user model based on the preferred test setting and the environment state so as to provide an updated user model; and updating the environment model based on the preferred test setting and the updated user model so as to provide an updated environment model.

[0128] The fitting agent 27 implemented in the accessory device 4 is configured to update the user model and / or the environment model, such as a plurality of user preference functions of the user model. In other words, the fitting agent is configured to update the user model based on the hearing device parameters and the environment data of the preferred test setting. Updating the user model may include updating the user preference model and the user response model based on the A test setting, the B test setting, the preferred test setting and the environment state. The fitting agent 27 / accessory device 4 may be configured to send the A test setting, the B test setting and the preferred test setting to the server device 5, the server device updates the user model and / or the environment model, and sends the updated user model and / or environment model to the fitting agent 27 / accessory device 4. Therefore, the fitting agent 27 / accessory device 4 may be configured to receive the updated model from the server device 5. In other words, the fitting agent 27 may be distributed on the accessory device 4 and one or more of the hearing device 2 and the server device 5. Thus, the fitting agent 27 may comprise a first part 27A implemented in the accessory device 4 , an optional second part 27B implemented in the server device, and an optional third part 27C implemented in the hearing device 2 (eg the processor 16 ).

[0129] In one or more example fitting agents including fitting agents 27, 27A, 27B, 27C, obtaining environmental data comprises obtaining audio data and determining environmental data based on the audio data. For example, the microphone 12 of the hearing device 2 may provide audio data representing sound received by the microphone 12, wherein the audio data may be sent to the accessory device 4, wherein the fitting agent 27, 27A of the accessory device 4 is configured to determine environmental data based on the audio data from the hearing device, and / or wherein the fitting agent 27C of the hearing device 2 is configured to determine environmental data or at least a portion of the environmental data based on the audio data from the microphone 12. The environmental data determined by the fitting agent 27C of the hearing device 2 is sent to the fitting agent 27A of the accessory device 4 for further processing.

[0130] In one or more example fitting agents including fitting agents 27, 27A, 27B, 27C, obtaining the environmental data includes, for example, obtaining contextual data from one or more sensors of the hearing device 2 and / or the accessory device 4, and determining the environmental data based on the contextual data. Determining the environmental data may be based on the audio data and the contextual data. In one or more example fitting agents including fitting agents 27, 27A, 27B, 27C, obtaining the contextual data includes, for example, obtaining the contextual data from one or more applications (e.g., a calendar application) of the accessory device 4.

[0131] In one or more example fitting agents including the fitting agents 27 , 27A, 27B, 27C, obtaining test settings for the hearing device including an A test setting and a B test setting includes: determining the B test setting based on an environment state.

[0132] In one or more example fitting agents including the fitting agent 27 , 27A, 27B, 27C, the fitting agent is configured to determine and update hearing device parameters of the hearing device based on the updated user model and optionally based on the updated environment model.

[0133] In an implementation including the accessory device 4 , the fitting agent 27 / accessory device 4 may be configured to send a control signal 30 to the hearing device 2 , the control signal 30 indicating the A test setting and the B test setting, thereby enabling the hearing device 2 to output a test signal accordingly.

[0134] The hearing device 2 (processor 16 ) is optionally configured to output an A test signal according to the A test setting via the receiver 22 , and to output a B test signal according to the B test setting via the receiver 22 .

[0135] The fitting agent 27, 27A, 27C (the hearing device 2 (processor 16) and / or the accessory device 4) is configured to detect user input of a preferred test setting indicating a preference for the A test setting or the B test setting, for example by detecting a user input on the user interface 20, or by detecting a user selection of one of the A virtual button 32 (preferred A test setting) and the B virtual button 34 (preferred B test setting) on ​​the user interface 24 of the accessory device 4.

[0136] It should be noted that the fitting agent 27, 27A, 27B, 27C may be configured to detect user input of a preferred test setting indicating a preference for the A test setting or the B test setting, for example by receiving a wireless input signal from the B accessory device (e.g. a smart watch including a user interface), thereby providing a more convenient user input, which in turn improves the user-friendliness of the fitting agent.

[0137] The user may initiate an update of the user model and / or environment model by pressing a start button 28 on the user interface of the accessory device 4. In other words, the fitting agent may detect a user input indicating a start and, in dependence on detecting the user input indicating a start, execute the method / update of the user model.

[0138] In an implementation comprising the accessory device 4 , the accessory device 4 may be configured to send a control signal 38 indicating hearing device parameters of a preferred test setup to the hearing device 2 , thereby enabling the hearing device to update the preferred hearing device parameters and apply them in the hearing device.

[0139] Figure 3 An interaction between a user 40 and a fitting agent 27 in an environment 41 is shown. The fitting agent 27 obtains environment data ENV_D indicative of a current environment 41, for example based on audio data and / or contextual data from one or more hearing devices worn by the user 40, the user providing user input to the fitting agent, accessory device microphones, sensors and / or applications.

[0140] For each interaction or trial indexed by index n, the fitting agent generates the A test setting, for example by controlling the hearing device to output an A test signal and a B test signal according to and indicative of the A test setting and the B test setting, respectively. and B test setup and present the user with (42) the test setup and Users evaluate these two test setups and and the fitting agent 27 receives and detects a user response 44 indicating a preferred test setting between the A test setting and the B test setting, n The fitting agent is based on the environment state ct 、r n , and to update the user model, and then update the environment model based on the updated user model.

[0141] The use of the terms "first", "second", "third", and "fourth", "a", "b", "c", etc. does not imply any particular order, but is intended to identify individual elements. Furthermore, the use of the terms "first", "second", "third", and "fourth", "a", "b", "c", etc. does not indicate any order or importance, but rather the terms "first", "second", "third", and "fourth", "a", "b", "c", etc. are used to distinguish one element from another. It should be noted that the words "first", "second", "third", and "fourth", "a", "b", "c", etc. are used here and elsewhere for labeling purposes only and are not intended to indicate any particular spatial or temporal order.

[0142] The memory may be one or more of a buffer, a flash memory, a hard drive, a removable medium, a volatile memory, a non-volatile memory, a random access memory (RAM), or other suitable devices. In a typical arrangement, the memory may include a non-volatile memory for long-term data storage and a volatile memory as a system memory of the processor. The memory may exchange data with the processor via a data bus. The memory may be considered a non-transitory computer readable medium.

[0143] The memory may be configured to store information in a portion of the memory.

[0144] Furthermore, the marking of a first element does not imply the presence of a second element, and vice versa.

[0145] It will be appreciated that the accompanying drawings include some modules or operations shown with solid lines, and some modules or operations shown with dotted lines. The modules or operations included in the solid lines are the modules or operations included in the most extensive exemplary implementations. The modules or operations included in the dotted lines are exemplary embodiments, which may be included in the modules or operations of the solid line exemplary embodiments or a part thereof, or other modules or operations outside the modules or operations of the solid line exemplary embodiments. It will be appreciated that these operations do not necessarily have to be performed in the order presented. In addition, it will be appreciated that not all operations need to be performed. The exemplary operations may be performed in any order and combination.

[0146] It should be noted that the word "comprising" does not necessarily exclude the presence of other elements or steps than those listed.

[0147] It should be noted that the word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements.

[0148] It should be noted that the word "based on" may be viewed as "as a function of" and / or "derived from". The terms "based on" and "as a function of" may be used interchangeably. For example, a parameter determined "based on" a data set may be viewed as a parameter determined "as a function of the data set". In other words, a parameter may be the output of one or more functions that have a data set as input.

[0149] It should also be noted that any reference signs do not limit the scope of the claims, that exemplary embodiments may be implemented at least partially by hardware and software, and that several "means", "units" or "devices" may be represented by the same item of hardware.

[0150] Various exemplary methods, devices, and systems described herein are described in the context of method steps, which in one aspect may be implemented by a computer program product (embodied in a computer-readable medium) comprising computer-executable instructions (such as program code) implemented by a computer in a network environment. Computer-readable media may include removable and non-removable storage devices, including but not limited to read-only memory (ROM), random access memory (RAM), compact disk (CD), digital versatile disk (DVD), etc. In general, program modules may include routines, programs, objects, components, data structures, etc. that perform specified tasks or implement specific abstract data types. Computer-executable instructions, associated data structures, and program modules represent examples of program code for performing the steps of the methods disclosed herein. A specific sequence of such executable instructions or associated data structures represents an example of corresponding actions for implementing the functions described in such steps or processes.

[0151] Although features have been shown and described, it is understood that the features are not intended to limit the claimed invention, and it will be apparent to those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the invention. Accordingly, the description and drawings should be regarded as illustrative rather than restrictive. The claimed invention is intended to cover all alternatives, modifications and equivalents.

[0152] List of reference numerals

[0153] 1 Hearing Device System

[0154] 2 Hearing devices

[0155] 4. Accessories

[0156] 5 Server Equipment

[0157] 6 Transceiver Module

[0158] 8 Antenna

[0159] 10 Transceiver

[0160] 11 Wireless connection between hearing aids and accessory devices

[0161] 11A Wireless connection between accessory equipment and server equipment

[0162] 12 First Microphone

[0163] 14 First microphone input signal

[0164] 16 Processors

[0165] 18 Electronic output signals

[0166] 20 User Interface

[0167] 22 Receiver

[0168] 24 User interface of accessory equipment

[0169] 26 Touch screen display

[0170] 27 Fitting Agent

[0171] 27A Fitting Agent Part I

[0172] 27B Part 2 of the Fitting Agent

[0173] 27C Part III of the Fitting Agency

[0174] 28 (virtual) start button

[0175] 30 Control signal indicating the setting of A and B tests

[0176] 32 A virtual button

[0177] 34 B Virtual Button

[0178] 38 Control signal indicating hearing device parameters for preferred test setup

[0179] 40 users

[0180] 41 Environment

[0181] 42 nth trial, test setup and

[0182] 44 User response to the nth trial

[0183] 46 n+1th trial, test setup and

[0184] 50 Environmental Models

[0185] 52 Preference Learning System

[0186] 54 User Model

Claims

1. A fitting agent for a hearing device system, the hearing device system comprising a hearing device worn by a hearing device user, wherein: The fitting agent includes one or more processors configured to: Initializing a user model, the user model comprising a plurality of user preference functions and associated user response distributions; Initialize the environment model; obtaining environmental data indicative of a current environment; using the environmental model, determining an environmental state based on the environmental data; Based on the environmental state, obtaining a test setting for the hearing device, the test setting comprising a test setting A and a test setting B; presenting the test setup to the hearing device user; obtaining user input of a preferred test setting indicating a preference for the A test setting or the B test setting; updating the user model based on the preferred test settings and the environmental status to provide an updated user model; as well as The environment model is updated based on the preferred test settings and one of the user model and the updated user model to provide an updated environment model.

2. The fitting agent according to claim 1, wherein: Updating the environment model comprises: determining an updated environment state based on one of the user model and the updated user model; and Based on the updated environmental state, the environmental model is updated.

3. The fitting agent according to any one of claims 1-2, wherein: Updating the environment model includes: The number of observed environment categories of the environment model is reduced.

4. The fitting agent according to any one of claims 1 to 3, wherein: Updating the environment model comprises: Increase the number of observed environment categories of the environment model.

5. The fitting agent according to any one of claims 1 to 4, wherein: The environment model is an infinite Gaussian mixture model.

6. The fitting agent according to any one of claims 1 to 5, wherein: Initializing the environment model includes: Obtain user profiles; and Based on the user profile, the environment model is initialized.

7. The fitting agent according to claim 6, wherein: The user profile includes one or more of age, gender, degree of hearing loss, and activity level, and Wherein, initializing the environment model is based on one or more of age, gender, hearing loss degree and activity level.

8. The fitting agent according to any one of claims 1 to 7, wherein: The environment model is a Hierarchical Dirichlet Process Hidden Markov Model (HDP-HMM).

9. The fitting agent according to claim 8, wherein: The environmental categories are modeled as latent random variables from a countably infinite category space.

10. The fitting agent according to any one of claims 8-9, wherein: Initializing the environment model includes: modeling the environment category using a Dirichlet process, and Here, each environment class generates a state-specific infinite Dirichlet-Gaussian mixture.

11. The fitting agent according to any one of claims 1 to 10, wherein: Updating the environment model comprises: Apply the viscous hierarchical Dirichlet process hidden Markov model.

12. The fitting agent according to any one of claims 1 to 11, wherein: The environment state is an environment identifier indicating the current environment.

13. A fitting agent according to any one of claims 1 to 12, wherein: The environment state is a probability distribution for the current environment.

Images