| CPC G16H 20/30 (2018.01) [G16H 10/60 (2018.01); G16H 40/67 (2018.01)] | 8 Claims |
|
1. A cloud-based hearing aid management system comprising:
a user device comprising:
a user database configured to store at least one piece of audiometry information;
a user processor connected to the user database and configured to access information in the user database, the user processor includes a user model, which is a neural network model, and the user processor uses the information in the user database to train the user model suitable for a user by using federated learning,
wherein the user model includes a forgetting valve, an input valve, and an output valve connected in sequence, and the user model calculates the prescription suitable for the user based on the following equations:
ft=σ(Wf·[ht−1,xt]+bf),
it=σ(Wi·[ht−1,xt]+bi),
Ct=tanh(WC·[ht−1,xt]+bC),
Ct=ft*Ct−1+it*Ct,
Ot=σ(WO·[ht−1,xt]+b0), and
ht=Ot+*tanh(Ct),
wherein ft represents a result of the forgetting valve, σ represents the Sigmoid function, ht−1 represents a weight of a hidden layer of a previous Cell, ht represents a weight of a hidden layer of a new Cell, xt represents an input vector, Wf·[ht−1,xt] represents a weight of the forgetting valve including the input vector and the previous hidden layer, bf represents an offset value of the forgetting valve, it represents a result of the input valve, Wi·[ht−1,xt] represents a weight of the input valve including the input vector and the previous hidden layer, bi represents an offset value of the input valve, Ct represents a new vector, tanh represents the tanh's activation function, WC·[ht−1,xt] represents a weight of the new vector including the input vector and the previous hidden layer, bC represents the new vector, Ct represents a state of the new Cell, Ct−1 represents a state of the previous Cell, Ot represents a result of the output valve, WO·[ht−1,xt] represents a weight of the output valve including the input and the previous hidden layer, and bO represents an offset value of the output valve;
a user output unit connected to the user processor, wherein the user processor is configured to extract the at least one piece of audiometry information in the user database and output it through the user output unit;
a user input unit connected to the user processor, wherein the user processor configured to generate an audiogram based on a user's input received by the user input unit indicating that if the at least one piece of audiometry information can be heard; and
a user signal transceiver connected to the user processor, wherein the user processor is configured to output the audiogram through the user signal transceiver;
a management server, connected to the user device, comprising:
a management signal transceiver connected to the user signal transceiver and configured to receive the audiogram;
a management processor connected to the management signal transceiver, wherein the management processor is configured to receive and convert the audiogram into a hearing parameter, and extract a hearing-loss characteristic value from the audiogram, the management processor further includes a cloud model, which is a is a neural network model, the management processor receives the user model fed back by the user device for training the cloud model; and
a management database connected to the management processor and configured to store a hearing aid recommendation table,
wherein the management processor is configured to incorporate the hearing-loss characteristic value into the hearing aid recommendation table to generate corresponding hearing aid recommendation information for the cloud model of the management processor to generate hearing aid adjustment parameters, and the hearing aid recommendation table comprises hearing-loss characteristic values in different numerical ranges, and models of hearing aids corresponding to hearing-loss characteristic values in different numerical ranges, and
wherein the management database is further configured to store a hearing prescription conversion formula and a hearing aid parameter adjustment table, and the management processor is configured to incorporate the audiogram into the hearing prescription conversion table to generate hearing prescription information, and then incorporate the hearing prescription information into the hearing aid parameter adjustment table to generate hearing aid adjustment parameters, and the management processor calculates the prescription suitable for the user based on the following equations:
Δt=formulaground−formulapred
δWgates=FW(Wgates,Δt)
δbgates=Fb(bgates,Δt)
δWc=Fw(Wc,Δt)
δbc=Fb(bc,Δt)
wherein, formulaground represents the optimal adjustment prescription finally satisfied by the user, formulapred represents the optimal adjustment prescription predicted by the model, Δt represents an error, FW represents a function for updating the weight of the model, Fb represents a function for updating the offset value of the model, Wgates represents the weights of the forgetting valve, the input valve, and the output valve, δWgates represents the new weights of the forgetting valve, the input valve, and the output valve after update, bgates represents the offset values of the forgetting valve, the input valve, and the output valve, δbgates represents the new offset values of the forgetting valve, the input valve, and the output valve after update, Wc represents the weight of the cell, δWc represents the new weight of the cell after update, bc represents the offset value of the cell, and δbc represents the new offset value of the cell; and
a hearing aid, connected to the user device, wherein the management processor sends the hearing aid adjustment parameters to the user signal transceiver and controls the user signal transceiver to transmit the hearing aid adjustment parameter to the hearing aid to adjust sound parameters generated by the hearing aid.
|