Information processing method and information processing system

Abstract

An information processing method for an information processing system (1) according to the present disclosure comprises a processing sound generation step and an adjustment step. In the processing sound generation step, a processing sound is generated by acoustic treatment using a parameter for changing the sound collecting function or the assistive listening function of an output sound unit. In the adjustment step, the output sound unit is adjusted by the parameter used for the acoustic treatment and a parameter selected on the basis of feedback on the processing sound outputted from the output sound unit.

Description

Technical Field

[0006] , ,

[0005] , , ,

[0007] , ,

[0001] The present disclosure relates to an information processing method and an information processing system.

Background Art

[0002] There is a device that allows a user to listen to environmental sounds in a suitable manner by adjusting parameters of an external sound capture function using a head-mounted acoustic device such as a hearing aid, a microphone, and earphones (see, for example, Patent Document 1).

[0003] Hearing aids require adjustment work according to an individual's hearing characteristics and use cases. For this reason, generally, an expert has adjusted parameters while counseling a user of a hearing aid.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, when adjusting parameters by a person such as an expert, there is a problem that the ability to adjust depends on the experience of the person who adjusts the parameters.

[0006] Therefore, the present disclosure proposes an information processing method and an information processing system that can appropriately adjust parameters of a hearing aid without being influenced by a person's experience.

Means for Solving the Problems

[0007] The information processing method of the information processing system relating to this disclosure includes a processed sound generation step and an adjustment step. The processed sound generation step generates a processed sound by acoustic processing using parameters that change the sound collection function or hearing aid function of the sound output unit. The adjustment step adjusts the sound output unit using parameters selected based on the parameters used in the acoustic processing and feedback to the processed sound output from the sound output unit. [Brief explanation of the drawing]

[0008] [Figure 1] This figure shows the basic learning model of this disclosure. [Figure 2] This figure shows a schematic configuration of an information processing system according to an embodiment of the present disclosure. [Figure 3] This figure shows an example of a deep neural network according to the embodiments of this disclosure. [Figure 4] This figure shows an example of a deep neural network according to the embodiments of this disclosure. [Figure 5] This figure shows a reward prediction unit according to an embodiment of the present disclosure. [Figure 6] This is an explanatory diagram illustrating the operation of an information processing system according to an embodiment of this disclosure. [Figure 7] This is an explanatory diagram illustrating the operation of an information processing system according to an embodiment of this disclosure. [Figure 8A] This is an explanatory diagram of a user interface according to an embodiment of the present disclosure. [Figure 8B] This is an explanatory diagram of a user interface according to an embodiment of the present disclosure. [Figure 9] This is a schematic diagram illustrating the adjustment system according to the embodiment of this disclosure. [Figure 10] This flowchart shows an example of a process performed by the information processing system according to the embodiment of this disclosure. [Figure 11] This flowchart shows an example of a process performed by the information processing system according to the embodiment of this disclosure. [Figure 12] This is an explanatory diagram of a user interface according to an embodiment of the present disclosure. [Figure 13] This is a diagram showing the configuration of a system including an external communication device and a hearing aid body according to an embodiment of the present disclosure. [Figure 14] This is a diagram showing an image of feedback acquisition according to an embodiment of the present disclosure. [Figure 15] This is an operation explanatory diagram of an information processing system according to an embodiment of the present disclosure. [Figure 16] This is a diagram showing the configuration of an external communication device including a user's situation estimator according to an embodiment of the present disclosure. [Figure 17] This is a flowchart showing an example of a process executed by an information processing system according to an embodiment of the present disclosure. [Figure 18] This is a diagram showing the configuration of a data aggregation system according to an embodiment of the present disclosure. [Figure 19] This is a diagram showing another configuration example of an adjustment system according to an embodiment of the present disclosure.

Embodiments for Carrying Out the Invention

[0009] Hereinafter, embodiments of the present disclosure will be described in detail based on the drawings. In each of the following embodiments, the same parts are denoted by the same reference numerals, and redundant descriptions are omitted.

[0010] [1. Overview of Information Processing System] The information processing system according to the present embodiment is a device that automatically or semi-automatically adjusts parameters (hereinafter, also referred to as "fitting") for changing the hearing function of an output device such as a hearing aid, a microphone, or an earphone having an external sound capture function. Hereinafter, the case where the information processing system performs fitting of a hearing aid will be described, but the target of parameter adjustment may be another output device such as a microphone or an earphone having an external sound capture function.

[0011] The information processing system performs hearing aid fitting by reinforcement learning, which is an example of machine learning. The information processing system includes an agent that asks questions to collect data for obtaining a method for predicting "reward" in reinforcement learning.

[0012] The agent conducts an A / B test on the hearing aid wearer (hereinafter referred to as "user"). The A / B test is a test in which the user is made to listen to the voice of A and the voice of B, and the user is asked to answer which of the voices of A or B is preferable. Note that the sounds to be listened to by the user are not limited to two types of A and B, and may be three or more types of voices.

[0013] As a method of answering the A / B test, for example, a UI (user interface) is used. As the UI, for example, buttons for selecting A or B are displayed on a smartphone or a smartwatch, and the user is made to select A or B by button operation. The UI may display a button for selecting "no difference between A and B".

[0014] Also, the UI may be a button that returns feedback only when the voice of B (output signal) with new parameters is more appropriate than the voice of A as an output signal with original parameters. Also, the UI may be configured to receive the user's answer by an action such as a user head shaking operation.

[0015] In addition, the information processing system can also collect the voices before and after adjustment by the user from electrical products (for example, smartphones and televisions) around the user as data, and perform reinforcement learning based on the collected data.

[0016] As a method for obtaining data for reward prediction other than the A / B test, for example, when an operation involving sound adjustment is performed, the sound and parameters before correction and the sound and parameters after correction are obtained and used as data for learning a reward predictor.

[0017] Furthermore, when conducting A / B testing, the information processing system might, for example, display an avatar agent, such as a person or character, on the UI, and have the agent act as an audiologist, interacting with the user while fitting the hearing aid.

[0018] [2. Background] Hearing aids employ a wide variety of signal processing techniques, but the most representative is the "compressor (non-linear amplification)" process. Therefore, unless otherwise specified, the following explanation will focus on adjusting the parameters of the compressor process.

[0019] In typical hearing aids, compressor adjustments are performed by an audiologist at a hearing aid store or similar location. The audiologist first measures the user's hearing and obtains an audiogram. Next, the audiologist inputs the audiogram into a fitting formula (e.g., NAL-NL, DSL, etc.) to obtain recommended compressor adjustment values.

[0020] Afterward, the audiologist has the user wear a hearing aid with the recommended compressor settings applied, and has them listen to sounds on the spot, asking for their feedback. If the user expresses dissatisfaction, the audiologist fine-tunes the compressor settings based on their knowledge.

[0021] However, there are several challenges when hearing aid fitting is performed by an audiologist. For example, the labor costs associated with providing support from an audiologist are high. Also, the quality of the fitting depends heavily on the experience of both the person making the adjustment and the person being adjusted, and satisfactory adjustments may not always be achieved. Furthermore, with infrequent adjustments, there are limitations to making fine-tuned adjustments. In addition, it is difficult to address users' dissatisfaction with their hearing in a timely manner.

[0022] Therefore, in this embodiment, we propose an information processing system and information processing method that can suitably adjust the parameters of a hearing aid without being influenced by human experience, by adjusting the parameters of the hearing aid using an information processing system without the intervention of an audiologist.

[0023] Reinforcement learning is a method that achieves this objective. Reinforcement learning is a method that "determines what action to take in order to maximize the total amount of rewards that can be obtained in the future."

[0024] Applying typical reinforcement learning to compressor tuning, the basic learning model can be realized with the configuration shown in Figure 1. In this case, the state s in reinforcement learning becomes an acoustic signal (processed sound) processed using certain parameters. The agent becomes an automatic parameter adjustment unit that selects one action a (= compressor parameter setting value) based on the input state at any given time.

[0025] Furthermore, in reinforcement learning, the environment processes the audio signal using the compressor parameter a selected by the agent to obtain s'. In addition, the following reward is obtained. The reward is a score r(s',a,s) that represents how much the user likes the parameter changes made by the agent.

[0026] The problem in reinforcement learning is to obtain a policy π(a|s) that maximizes the total reward obtained when an agent and its environment interact (exchanges of rewards, actions, and states) for a certain length of time. This problem can be solved using general reinforcement learning methodologies if the reward function r can be appropriately designed.

[0027] However, the extent to which individual users will like parameter changes is unknown, and this problem cannot be solved with the above approach. This is because it is impractical for humans to reward every single trial in a learning process involving a vast number of trials.

[0028] [3. Outline of the Information Processing System] As shown in Figure 2, the information processing system 1 according to this embodiment comprises an adjustment unit 10 and a processing unit 20. The processing unit 20 includes an environment generation unit 21. The environment generation unit 21 has the function of generating processed sound by acoustic processing (sound collector signal processing) using parameters that change the hearing function of the hearing aid and outputting the processed sound from the hearing aid.

[0029] The adjustment unit 10 acquires the parameters used in the acoustic processing and the user's feedback response to the processed sound, learns a method for selecting parameters suitable for the user through machine learning, and adjusts a hearing aid, which is an example of a sound output unit, using the parameters selected according to that selection method.

[0030] The adjustment unit 10 comprises an agent 11 and a reward prediction unit 12. The agent 11, as shown in Figure 1, learns from machine learning how to select parameters suitable for the user based on the input processing sound and reward, and outputs the parameters selected according to that selection method to the processing unit 20.

[0031] The processing unit 20 outputs the processed sound, which has been acoustically processed according to the input parameters, to the agent 11 and the reward prediction unit 12. Furthermore, the processing unit 20 outputs the parameters used for acoustic processing to the reward prediction unit 12.

[0032] The reward prediction unit 12 performs machine learning to predict rewards on behalf of the user based on sequentially input processing sounds and parameters, and outputs the predicted reward to agent 11. This allows agent 11 to suitably adjust the parameters of the hearing aid without the intervention of an audiologist and without the user having to perform a vast number of A / B tests.

[0033] [4. Learning and Adjustment Process] The reward prediction unit 12 acquires an audio signal for evaluation. In this embodiment, a dataset of input audio (processed sound) used in parameter adjustment is predetermined, and these processed sounds and the parameters used for acoustic processing of the processed sounds are randomly input to the reward prediction unit 12. The reward prediction unit 12 predicts a reward from the input processed sounds and parameters and outputs it to the agent 11.

[0034] Agent 11 selects an appropriate action (parameter) for the user based on the input reward and outputs it to the processing unit 20. The processing unit 20 obtains (updates) parameters θ1 and θ2 based on the action obtained from Agent 11.

[0035] In this embodiment, the signal processing to be adjusted is a 3-band multiband compressor process. The compression rate for each band can take three values ​​from a reference value, for example, -2, +1, and +4.

[0036] The reference value is the compression rate value calculated from the audiogram using a fitting formula. For example, if we consider a case with 3 patterns × 3 bands, the output from agent 11 will take on 9 values. The processing unit 20 applies signal processing for each parameter to the acquired audio.

[0037] The goal of this parameter adjustment step is to enable the reward prediction unit 12 and agent 11 to learn from the audio input that is received moment by moment, so that they can select the parameter set that the user will like best from among nine possible parameter sets for a given input and perform audio processing.

[0038] In the learning process including the reward prediction unit 12, the reward prediction unit 12 is first trained using supervised learning as preparation before reinforcement learning. Since it may be difficult for many users to listen to a single sound source and evaluate it absolutely, we consider an evaluation task in which the user is given two sounds, A and B, to listen to and respond to which one is easier to hear.

[0039] Figures 3 and 4 show specific examples of deep neural networks that learn the behavior of user responses in this task. The first and second input audio shown in Figure 3 are obtained by processing a single audio signal using two compression parameter sets θ1 and θ2, respectively. Note that the first and second input audio shown in Figure 3 may be preprocessed by converting them into short-time Fourier transform amplitude spectra and log-mel spectra, etc.

[0040] The first and second input audio signals are fed into the shared network shown in Figure 4. The first and second outputs from the shared network are fed into a fully connected layer, coupled, and then fed into the softmax function.

[0041] The output of the reward prediction unit 12 shown in Figure 3 is the probability that the first input audio is preferable to the second input audio. The following λ is used as training data for the output. λ=(λ1,λ2)=(1,0) indicates that the first input audio is preferable, λ=(λ1,λ2)=(0,1) indicates that both are within an acceptable range and no difference is perceived, and λ=(λ1,λ2)=(0.5,0.5) indicates that both are outside an acceptable range, while λ=(λ1,λ2)=(0,0) does not need to be used for training.

[0042] In this case, the network in Figure 3 can be optimized by learning to minimize the cross-entropy L = -Σ(λ1logP + λ2(1-P)) of the training data, where P is the output of the network. Also, the parameters θ1 and θ2 are randomly generated from possible choices. This is because, before the reinforcement learning process is run, appropriate input cannot be obtained from agent 11.

[0043] Unlike typical supervised learning model building use cases, the above learning process requires learning the preferences of individual users. Therefore, it is necessary to acquire data over a certain period of time after the purchase of a hearing aid. However, as will be described later, the reward prediction unit 12 has opportunities to be updated further, so it is not necessarily required that the learning process be fully completed at this point.

[0044] Next, we will explain standard reinforcement learning. Using the reward prediction unit 12 obtained through the above learning, we repeatedly update agent 11 using typical reinforcement learning. First, the objective function in reinforcement learning is expressed by the following equation (1).

number

number

number

number

[0045] The agent update in reinforcement learning is given as follows: 1. Initialize the policy π using, for example, a uniform distribution. 2. Run the following loop: (a) Determine the action (=compression parameters) according to the current policy, and calculate the reward value for the current state using the reward predictor (reward prediction unit 12) shown in Figure 5. Then, input the action (=compression parameters) to the environment and obtain the next state. After that, (b) estimate the action value function = Q for the next state, and (c) update the policy using the estimated Q.

[0046] While various reinforcement learning methods exist depending on how (i) and (iii) above are implemented, Q-learning will be used as an example here. Note that Q-learning is not the only reinforcement learning method that can achieve (i) and (iii) above.

[0047] In Q-learning, the Q-value for the next step, based on the definition of Q(s,a;Φ), is given by the following equation (5):

number

number

[0048] Figure 6 shows the operation of the information processing system 1 in this step. As shown in Figure 6, the system determines the action (=compression parameters) according to the current policy and outputs the parameters to the processing unit 20. The processing unit 20 processes the audio signal for learning based on the input parameters and outputs the processed sound to the agent 11. The processing unit 20 also outputs the pair of processed sounds (first input audio and second input audio) and the parameters to the reward prediction unit 12.

[0049] The reward prediction unit 12 estimates the reward from the processed sound pair and parameters, and outputs the estimated reward to the agent 11. Based on the input reward, the agent 11 determines the optimal action (=compression parameters) and outputs the parameters to the processing unit 20. The information processing system 1 updates the agent 11 and the reward prediction unit 12 through reinforcement learning while repeating this operation.

[0050] Furthermore, if the information processing system 1 receives feedback from the user, it updates the reward prediction unit 12 asynchronously. If the information processing system 1 has updated the agent 11 to a certain extent and it is expected that the action value function and policy have reached a reasonable value, it can obtain further user feedback and update the reward prediction unit 12.

[0051] In this case, unlike the first step, the parameters θ1 and θ2 used to generate the first and second input voices may be such that θ1 is the parameter from the previous step and θ2 is the parameter obtained from agent 11 in the current step.

[0052] Figure 7 shows the operation of the information processing system 1 in this step. As shown in Figure 7, the information processing system 1 presents a pair of processed sounds output from the processing unit to the user via the user interface 30. The information processing system 1 then outputs the user's feedback (response: which sound is better) to the processed sounds, along with the pair of processed sounds, to the reward prediction unit 12. The other operations are the same as those shown in Figure 6.

[0053] [5. User Interface] Next, an example of a user interface related to this disclosure will be described. The user interface is implemented, for example, by the display and operation unit (e.g., touch panel display) of an externally connected device such as a smartphone, smartwatch, or personal computer.

[0054] External devices come pre-installed with an application program for adjusting hearing aid parameters (hereinafter referred to as the "adjustment app"). Some functions for adjusting hearing aid parameters may also be implemented as functions of the external device's operating system (OS). Users launch the adjustment app by operating the external device when purchasing a hearing aid or when they are dissatisfied with the hearing aid's behavior.

[0055] When the external device launches the adjustment application, it displays, for example, the user interface 30 shown in Figure 8A. The user interface 30 includes a display unit 31 and an operation unit 32. The display unit 31 displays an avatar 33 that speaks adjustment processing sounds.

[0056] The control unit 32 includes sound output buttons 34 and 35 and 1-4 keys 36, 37, 38, and 39. When the user taps the sound output button 34, the avatar 33 speaks the first input voice, A, and when the user taps the sound output button 35, the avatar 33 speaks the second input voice, B.

[0057] The user interface 30 outputs feedback to the reward prediction unit 12, indicating "Audio A is easy to hear" when key 1 36 is tapped, and "Audio B is easy to hear" when key 2 37 is tapped.

[0058] Furthermore, when the 3 key 38 is tapped, the user interface 30 outputs feedback to the reward prediction unit 12 stating, "I don't perceive any difference between voices A and B; both are within acceptable limits," and when the 4 key 39 is tapped, it outputs feedback stating, "I don't perceive any difference between voices A and B; both are unpleasant." In this way, the user interface 30 allows users to easily conduct A / B tests through interaction with the avatar 33, regardless of their location.

[0059] The external device may display the user interface 30 shown in Figure 8B. In the example shown in Figure 8B, the display unit 31 displays an avatar 33a of an audiologist, who is a specialist in hearing aid fitting.

[0060] When the adjustment app is launched, Avatar 33a acts as a facilitator, guiding the hearing aid adjustment process with questions such as, "Which do you prefer, A or B?" or "How about C?". In this way, the adjustment app can present information and options in a conversational format, as if a virtual audiologist agent, either live-action or animated, were remotely fitting the hearing aid to the user.

[0061] By using such a user interface 30, it is expected that the user's stress from repeatedly performing monotonous tests and the user's stress from adjustment failures, such as when the system suggests parameter settings that produce an undesirable sound, can be alleviated.

[0062] Furthermore, the user interface 30 shown in Figure 8B displays a slider 36a instead of the 1-4 keys 36, 37, 38, and 39. This allows the user to respond with a continuous value between 0 and 1 as their preference for the voice, rather than a 0 or 1 response, by using the slider 36a on the app.

[0063] For example, if slider 36a is positioned midway between A and B (0.5), the user will not perceive any difference between A and B and will find both acceptable. If slider 36a is positioned closer to B (0.8), the user will be able to answer something like, "I prefer B."

[0064] Furthermore, the response method for A / B testing using the adjustment app may be an audio response such as "I like A" or "I like B." Also, for example, if the audio for A is output first, followed by the audio for B, the system may be configured to respond by nodding its head to indicate whether the changed parameters are acceptable. Additionally, if there is no vertical nod indicating acceptance within a predetermined time (e.g., 5 seconds) after the sound is output, it may be considered a rejection.

[0065] Up to this point, we have described examples of hearing aid adjustments and user feedback acquisition using external devices, but hearing aid adjustments and feedback acquisition may also be performed without using external devices. For example, the hearing aid may output voice A, voice B, and guidance voice, and the user may input feedback using physical keys, contact sensors, proximity sensors, accelerometers, or microphones provided on the hearing aid itself, in accordance with the guidance voice.

[0066] [6. Overview of the Adjustment System] Next, an overview of the coordination system related to this disclosure will be described. Here, we will describe the case where the external cooperating device has the functions of the information processing system 1. As shown in Figure 9, the external cooperating device 40 is connected to the left ear hearing aid 50 and the right ear hearing aid 60 via wired or wireless communication.

[0067] The system comprises an adjustment unit 10, a left ear hearing processing unit 20L, a right ear hearing processing unit 20R, and a user interface 30. The adjustment unit 10, the left ear hearing processing unit 20L, and the right ear hearing processing unit 20R include a microcomputer with a CPU (Central Processing Unit), ROM (Read Only Memory), RAM (Random Access Memory), and various circuits.

[0068] The adjustment unit 10, the left ear hearing processing unit 20L, and the right ear hearing processing unit 20R function by having the CPU execute the adjustment application stored in ROM, using RAM as a working area.

[0069] Furthermore, the adjustment unit 10, the left ear hearing assistance processing unit 20L, and the right ear hearing assistance processing unit 20R may be partially or entirely composed of hardware such as an ASIC (Application Specific Integrated Circuit) or FPGA (Field Programmable Gate Array).

[0070] As mentioned above, the user interface 30 is implemented, for example, by a touch panel display. The left ear hearing aid 50 includes a left ear acoustic output unit 51. The right ear hearing aid 60 includes a right ear acoustic output unit 61.

[0071] At least one of the left ear hearing aid 50 and the right ear hearing aid 60 may be equipped with an acoustic input unit (not shown) for collecting ambient sounds, which may consist of a microphone or the like. The acoustic input unit may also be provided in an external device 40 or other device that is wired or wirelessly connected to the left ear hearing aid 50 and the right ear hearing aid 60. The left ear hearing aid 50 and the right ear hearing aid 60 perform compression processing based on the ambient sounds acquired by the acoustic input unit. The ambient sounds acquired by the acoustic input unit may be used by the left ear hearing aid 50, the right ear hearing aid 60, or the external device 40 for noise suppression, beamforming, or voice instruction input functions.

[0072] The adjustment unit 10 includes an agent 11 and a reward prediction unit 12 (see Figure 2), and outputs parameters to the left ear hearing processing unit 20L and the right ear hearing processing unit 20R. The left ear hearing processing unit 20L and the right ear hearing processing unit 20R generate processed sounds by acoustic processing using the input parameters and output them to the left ear hearing aid 50 and the right ear hearing aid 60, respectively.

[0073] The left ear acoustic output unit 51 and the right ear acoustic output unit 61 output processed sounds input from the external connected device 40. The user interface 30 receives feedback from the user who listens to the processed sounds (which sound, A or B, is better) and outputs it to the adjustment unit 10. Based on the feedback, the adjustment unit 10 selects more appropriate parameters and outputs them to the left ear hearing processing unit 20L and the right ear hearing processing unit 20R.

[0074] The external device 40 repeats this operation, and once it has determined the optimal parameters, it sets the parameters of the left ear hearing aid 50 using the left ear hearing processing unit 20L and sets the parameters of the right ear hearing aid 60 using the right ear hearing processing unit 20R, thereby completing the parameter adjustment.

[0075] [7. Processes executed by the information processing system] Next, an example of the processing performed by the information processing system 1 will be described. As shown in Figure 10, when the adjustment application is launched, the information processing system 1 first determines whether or not there is a learning history (step S101).

[0076] If the information processing system 1 determines that there is a learning history (step S101, Yes), it moves the process to step S107. If the information processing system 1 determines that there is no learning history (step S101, No), it randomly selects a file from the evaluation audio data (step S102), randomly generates parameters θ1 and θ2, generates processed sounds A and B using those parameters, and plays them to conduct an A / B test (step S104).

[0077] Subsequently, the information processing system 1 obtains user feedback (for example, the input of keys 1, 2, 3, and 4 as shown in Figure 8A) (step S104) and determines whether the A / B test has been completed 10 times or not (step S105).

[0078] If the information processing system 1 determines that the process has not been completed 10 times (step S105, No), it moves the process to step S102. If the adjustment unit 10 determines that the process has been completed 10 times (step S105, Yes), it updates the reward prediction unit 12 using the data from the most recent 10 times (step S106).

[0079] Next, the information processing system 1 randomly selects a file from the evaluation data (step S107), randomly generates parameters θ1 and θ2, and generates processed sounds A and B using those parameters, and plays them to perform an A / B test (step S108).

[0080] Subsequently, the information processing system 1 obtains user feedback (for example, the 1, 2, 3, and 4 key inputs shown in Figure 8A) (step S109) and updates agent 11 (step S110).

[0081] Next, the information processing system 1 determines whether or not the A / B test has been completed 10 times (step S111). If the information processing system 1 determines that it has not been completed 10 times (step S111, No), it moves the process to step S107.

[0082] Furthermore, if the adjustment unit 10 determines that the process has been completed 10 times (step S111, Yes), it updates the reward prediction unit 12 using the data from the most recent 10 times (step S112) and determines whether the process from steps S106 to S112 has been completed twice (step S113).

[0083] If the information processing system 1 determines that the process has not been completed twice (step S113, No), it moves the process to step S106. Alternatively, if the information processing system 1 determines that the process has been completed twice (step S113, Yes), it terminates the parameter adjustment.

[0084] Furthermore, since inputting feedback for each A / B test is time-consuming, the information processing system 1 can also perform a simplified process as shown in Figure 11. Specifically, as shown in Figure 11, the information processing system 1 can also perform a process that omits steps S109, S112, and S113 of the process shown in Figure 10.

[0085] However, if the information processing system 1 executes the process shown in Figure 11, and the reward prediction unit 12 has not learned sufficiently, the output may deviate from the actual user's preferences, and learning may not be successful. Therefore, a restriction may be imposed so that the process shown in Figure 11 cannot be executed continuously.

[0086] [8. Other Examples] The embodiments described above are examples, and various modifications are possible. For example, the information processing method according to this disclosure can be applied not only to compression, but also to noise suppression, feedback cancellation, and automatic adjustment of parameters for specific direction emphasis by beamforming.

[0087] When adjusting multiple types of parameters, the information processing system 1 can learn multiple signal processing parameters in a single reinforcement learning process, or it can run the reinforcement learning process in parallel for each parameter subset. For example, the information processing system 1 can perform separate A / B testing and learning processes for noise suppression and A / B testing and learning processes for compression parameters.

[0088] Furthermore, the information processing system 1 can increase the number of conditional variables during learning. For example, it is possible to provide separate tests, separate agents 11, and separate reward prediction units 12 for each of several scenes and learn them individually.

[0089] [8-1. Obtaining indirect user feedback] Information processing system 1 can also obtain indirect user feedback through an app that adjusts some parameters of the hearing aid.

[0090] Some hearing aids offer a function to adjust certain parameters of the hearing aid directly or indirectly using, for example, a smartphone. Figure 12 shows an example of a user interface 30 that allows adjustment of certain parameters of a hearing aid.

[0091] As shown in Figure 12, the user interface 30 includes a slider 36b for adjusting the volume, a slider 37b for adjusting the 3-band equalizer, and a slider 38b for adjusting the strength of the noise suppression function.

[0092] Figure 13 shows the configuration of the system including the external connectivity device and the hearing aid unit. As shown in Figure 13, the external connectivity device 40 includes input audio buffers 71, 75, feedback acquisition units 72, 76, parameter buffers 73, 77, parameter control unit 78, user feedback DB (database) 74, and user interface 30. The parameter control unit 78 provides the functions of the information processing system 1.

[0093] The left ear hearing aid 50 includes a left ear acoustic output unit 51, a left ear acoustic input unit 52, and a left ear hearing processing unit 53. The right ear hearing aid 60 includes a right ear acoustic output unit 61, a right ear acoustic input unit 62, and a right ear hearing processing unit 63.

[0094] The left ear hearing aid 50 and the right ear hearing aid 60 transmit input audio to the external device 40. The external device 40 stores the received audio along with a timestamp in input audio buffers (for example, circular buffers of 60 seconds each for the left and right ears) 71, 75. This communication may be ongoing or may be initiated by the launch of the adjustment application or by user instruction.

[0095] When a user action is detected to control a parameter change, the parameter before the change is stored in parameter buffers 73 and 77 along with a timestamp. Subsequently, when the end of the parameter change is detected, the modified parameter is also stored in parameter buffers 73 and 77 along with a timestamp.

[0096] Each ear's parameter buffers 73 and 77 can store at least two parameter sets, one before and one after the change. The end of the parameter change can be detected, for example, when there is no operation for a predetermined time (e.g., 5 seconds). This predetermined time can be specified by the user, or the user can notify the system when the adjustment is complete.

[0097] Once the parameter adjustment is complete, the buffered audio and parameter sets are input to the feedback acquisition units 72 and 76. Figure 14 shows an image of the feedback acquisition process. As shown in Figure 14, two sets of feedback data can be obtained from the buffered audio input (before adjustment and after adjustment) and parameters (before adjustment and after adjustment).

[0098] Specifically, if a user listens to the processing sound of parameter θ1, then manually adjusts the parameter, and then listens to the processing sound of parameter θ2, it can be inferred that the processing sound of parameter θ2 is more to the user's liking than the processing sound of parameter θ1. In other words, it can be inferred that the user prefers parameter θ2 to parameter θ1.

[0099] Therefore, the feedback acquisition units 72 and 76 can label the first pair of processed sounds, A (processed sound A with parameter θ1 before adjustment) and B (processed sound B obtained by applying parameter θ2 to the input signal that is the source of the processed sound), with "I prefer B to A" and store them in the user feedback DB 74.

[0100] Furthermore, the feedback acquisition units 72 and 76 can label the first pair of processed sounds, A (processed sound A with adjusted parameter θ2) and B (processed sound B obtained by applying parameter θ1 to the input signal that is the source of the processed sound), with "I prefer A to B" and store them in the user feedback DB 74.

[0101] The parameter control unit 78 may immediately update the reward prediction unit 12 using the feedback stored in the user feedback DB 74, or it may update the reward prediction unit 12 using the feedback accumulated until some feedback data has been collected, or at regular intervals.

[0102] Thus, the adjustment unit 10 of the parameter control unit 78 uses machine learning to learn how to select parameters and how to predict rewards, based on the parameters before and after manual adjustment by the user, and the user's predicted response to the processed sound using those parameters.

[0103] In addition to the examples described here, the external device 40 can also acquire feedback data in a similar manner by using the sound before and after adjustment when sound adjustment operations are performed on sound-producing products such as televisions and portable players.

[0104] [8-2. Utilizing Additional Property Information] When adjusting hearing aid parameters, the preferred settings may differ depending on the user's situation, even for similar sound inputs. For example, during a meeting, an output that makes it easy to understand what is being said, even if the audio is slightly unnatural due to the side effects of signal processing, would be expected. Conversely, when relaxing at home, an output that minimizes sound quality degradation would be expected.

[0105] This means that the behavior of the policy and reward function in reinforcement learning differs depending on the user's situation. Therefore, one could consider including additional property information as a state to indicate "what situation the user is in."

[0106] Additional property information includes, for example, scene information selected by the user from the user interface 30 of the externally connected device 40, information entered by voice, the user's location information determined by GPS (Global Positioning System), the user's acceleration information detected by an acceleration sensor, and calendar information registered in an application program that manages the user's schedule, as well as combinations thereof.

[0107] Figure 15 shows the operation of the information processing system 1 when utilizing additional property information. As shown in Figure 15, the user uses the user interface 30 from the adjustment app to select "which scene adjustments they want to make."

[0108] In the previously described embodiment, the sound output from the environment generation unit 21 was randomly selected from all the sounds included in the evaluation data. In this embodiment, however, sounds that use environmental sounds that match the scene information are output from the evaluation data.

[0109] In this case, each audio data in the evaluation database needs to have metadata attached that indicates what kind of scene the sound is from. The reward prediction unit 12 and agent 11 also receive data indicating the user's situation, along with information on the processed sound and feedback.

[0110] The reward prediction unit 12 and agent 11 may be implemented with independent models that switch according to the input user situation, or they may be implemented as a single model that takes the user situation as input along with voice input.

[0111] Figure 16 shows the configuration of the external device 40a, which includes a user status estimator. The external device 40a differs from the external device 40 shown in Figure 13 in that it includes a sensor 79 and a collaboration application 80. The sensor 79 includes, for example, a GPS sensor or an accelerometer.

[0112] The linked application 80 includes, for example, calendar applications and social networking applications that include user status as text or metadata. The sensor 79, the linked application 80, and the user interface 30 input user status or information that can be used to estimate it to the feedback acquisition units 72, 76 and the parameter control unit 78.

[0113] The feedback acquisition units 72 and 76 use this information to classify the user's situation into one of the pre-defined categories, and add this classified information to the voice input and user feedback information before storing it in the user feedback DB 74.

[0114] The feedback acquisition units 72 and 76 may also detect scenes from the buffered audio input. The parameter control unit 78 selects appropriate parameters for each classified category using the machine learning-trained agent 11 and the reward prediction unit 12.

[0115] [8-3. Confidence (Weighting) of Feedback Data] In addition to the additional profile information described above, a confidence level may be added to each piece of feedback data. For example, when training the reward prediction unit 12, instead of inputting all data with uniform probability as training data, the data may be input in proportions according to the confidence level.

[0116] Regarding confidence levels, for example, when conducting A / B testing, the confidence level could be set to 1.0, while the indirect feedback (responses) obtained from smartphone adjustments, as described above, could be set to 0.5. In short, a predetermined value could be adopted depending on the source of the feedback data.

[0117] Alternatively, the level of confidence can be determined based on the surrounding environment during adjustments and the user's condition. For example, if the environment in which A / B testing is being conducted is noisy, the ambient noise may act as interference, potentially preventing the user from providing appropriate feedback.

[0118] Therefore, it is also possible to calculate the average equivalent noise level of ambient sound over several seconds, and set the confidence level to 0.5 if the average equivalent noise level is above the first threshold and below the second threshold which is higher than the first threshold; set the confidence level to 0.1 if it is above the second threshold and below the third threshold which is higher than the third threshold; and set the confidence level to 0 otherwise.

[0119] [8-4. Automatic fitting on the spot] In the embodiments described above, a use case was shown in which parameters are adjusted using the user interface 30 shown in Figure 12, and an example was described in which the information obtained therefrom is used for reward prediction. However, it is not possible to adjust all parameters of the hearing aid using the user interface 30 shown in Figure 12.

[0120] In the first place, there are use cases where manual adjustment of numerous parameters is complex and difficult for the user to perform, and therefore, automatic adjustment is used to make adjustments on a case-by-case basis. Therefore, information processing system 1 can combine manual parameter adjustment with automatic parameter adjustment.

[0121] In this case, the information processing system 1 executes the process shown in Figure 17, for example. Specifically, as shown in Figure 17, when the adjustment application is launched, the information processing system 1 first allows the user to perform manual adjustments (step S201), and then stores the adjustment results in the user feedback DB 74 (step S202).

[0122] Next, the information processing system 1 updates the reward prediction unit 12 (step S203) and determines whether the user wishes to perform further automatic adjustments (step S204). If the information processing system 1 determines that the user does not wish to perform further adjustments (step S204, No), it reflects the parameters before adjustment back into the hearing aid (step S212) and ends the adjustment.

[0123] Furthermore, if the information processing system 1 determines that the user desires it (step S204, Yes), it performs reinforcement learning by the reward prediction unit 12 (steps S107 to S111 shown in Figure 11) N times (N is an arbitrarily set natural number) (step S205).

[0124] Next, the information processing system 1 performs parameter updates by agent 11 and conducts A (before update) / B (after update) tests (step S206), stores the results in the user feedback DB 74 (step S207), and updates the reward prediction unit 12 (step S208).

[0125] Subsequently, the information processing system 1 determines whether the feedback is A (before update) or B (after update) (step S209). If the feedback is A (before update) (step S209, A), the information processing system 1 moves the process to step S204.

[0126] Furthermore, if the feedback is B (after update) (step S209, B), the information processing system 1 reflects the new parameters in the hearing aid and displays a message prompting confirmation of the adjustment effect in response to real voice input (step S210).

[0127] Subsequently, the information processing system 1 determines whether the user is satisfied or not (step S211). If it determines that the user is not satisfied (step S211, No), it proceeds to step S204. If the information processing system 1 determines that the user is satisfied (step S212, Yes), it terminates the adjustment.

[0128] [8-5. Utilization of adjustment information by audiologists] In the case of hearing aids, instead of relying entirely on automatic adjustment, there are use cases where users request adjustments from an audiologist. By adopting the following configuration, automatic parameter adjustments can be performed utilizing adjustment information from the audiologist.

[0129] The advantages of utilizing information from audiologists' adjustments are as follows. For example, from the perspective of hearing protection, the example given above showed "adding -2, +1, and +4 to the parameters for each band of the compressor, based on the base adjustment value." However, in actual use cases, a wider adjustment range may be necessary to achieve the desired effect. On the other hand, allowing the same adjustment range for all users would be problematic from the perspective of hearing protection.

[0130] Furthermore, from the perspective of getting used to hearing aids, users who are not accustomed to wearing hearing aids tend to prefer a lower amplification level than what the audiologist considers appropriate. Therefore, typically, a process is followed where the user gradually adjusts from their preference to the audiologist's appropriate level over time, allowing them to slowly get used to the sound of the hearing aid. Alternatively, some hearing aid stores may force users to use the level deemed appropriate by the audiologist.

[0131] To take advantage of these benefits, for example, if there is a clearly defined range of parameters that "must be this way," then the range of possible actions should be clearly defined. In the example above, it was stated that "for each band of the compressor, the parameters are added by -2, +1, +4 based on the base adjustment value," but this can be implemented by changing the set of values ​​from (-2, +1, +4) to (0, +2, +4, +6, +8, +10) or (-4, -2, 0, +2), etc. Note that the parameter settings can also be changed for each band. This approach is particularly effective from the perspective of hearing protection.

[0132] While it is not possible to determine a precise parameter range, in cases where "we want to incorporate elements that the audiologist deems good into the adjustments," it is advisable to configure a reward prediction unit 12 by the audiologist separately from the user's reward prediction.

[0133] For example, in a case where "the user strongly desires a compressor parameter of +5, and it's acceptable to set it there, but the audiologist predicts that there's a high probability that an appropriate value exists up to +4," a modified predicted reward like the one in equation (8) below is used. rtotal=ruser+raudi····(8)

[0134] Here, rtotal is the reward used for learning, and ruser is the output of the reward prediction unit 12. Raudi can use a function like raudi = -β / exp(+a(x-4))1, which gradually reduces the reward when the parameter value x exceeds +4. If you want to leverage the evaluation of the audiologist's implicit adjustment results, you can train raudi in the same way as ruser.

[0135] Alternatively, instead of implementing a special mechanism to incorporate adjustment results by an audiologist, the parameters before and after adjustment, obtained from in-store adjustments or remote fittings, as well as the processed sound used for listening tests to confirm the effects, may be stored in the User Feedback DB74 and used as reinforcement learning data.

[0136] [8-6. Examples of aggregating and utilizing data from multiple users] Up to this point, we have discussed cases where only individual data is used to adjust a user's hearing aids. However, it is also possible for the service provider to aggregate data from multiple users to improve the quality of the automatic adjustment function for each user.

[0137] This embodiment is based on the assumption that "users with similar personal profiles and hearing loss symptoms should have similar reward functions and preferred adjustment parameters." A schematic diagram of the system configuration of this embodiment is shown in Figure 18.

[0138] In Figure 18, within the external linked devices 4-1 to 4-N of each of the first to Nth users U-1 to U-N, countless pieces of feedback data are accumulated by using the adjustment functions described above.

[0139] This data, along with the user identifier, the identifiers of the hearing aids 5-1 to 5-N used to collect the feedback data, the parameters of agent 11 and reward prediction unit 12 in reinforcement learning, and the parameters of the adjusted hearing aids 5-1 to 5-N, are combined and uploaded to the feedback database 74a on the server.

[0140] External devices 4-1 to 4-N are directly connected to the WAN (Wide Area Network) and can upload data in the background, or they can transfer the data to another external device such as a PC and upload it from there. This feedback data is assumed to include property information as described in [8-2. Utilization of Additional Property Information].

[0141] The user feedback analysis processing unit 81, for example, uses information such as "native language, age group, and usage scene" as is, or performs clustering in a space where audiogram information is used as a feature vector (for example, k-means clustering) to classify users into a predetermined number of classes and then classifies the aggregated various information.

[0142] The information that characterizes the classification itself (for example, the property information itself, the average value of each class in the clustered audiogram), along with all or part of the classified feedback data and user data, or representative values ​​and statistics, are stored in the shared DB74b.

[0143] The representative values ​​may be the summation average for each classification in the audiogram feature space, or the data of the individual closest to the median. Alternatively, the reward prediction unit 12 and agent 11 may be retrained using feedback data from all classified users or a subset of users close to the median. The learning itself is performed by applying the method described in the above embodiment to the data of multiple users.

[0144] One specific use of the shared DB74b obtained in this way is data sharing with users who have just started using hearing aids. In the embodiment described above, the initial values ​​of the compressor parameters were calculated from the fitting formula based on the audiogram, but in this embodiment, instead, representative values ​​of the class classified based on the user profile, or the data of the nearest user within the same classification may be used as the initial values. The same applies not only to the initial values ​​of the adjustment parameters, but also to the initial values ​​of the agent 11 and the reward prediction unit 12.

[0145] The second specific application is its use in the tuning process. In addition to updating parameters based on actions output by agent 11, randomly selecting tuning parameters from the same user class at a predetermined frequency can be expected to prevent convergence to local optima and accelerate the discovery of better solutions.

[0146] [8-7. Other Configuration Examples of the Adjustment System] Figures 9, 13, and 16 show examples where the left and right hearing aids are equipped with independent input audio buffers, parameter buffers, and feedback acquisition units 72, 76. This is because many hearing aid users wear hearing aids in both ears, and the symptoms of hearing loss differ between the left and right ears, requiring independent compressor parameters for each.

[0147] If the user wears a hearing aid in only one ear, the system can be implemented with a configuration for one ear. Among the parameters for hearing aid signal processing other than the compressor, some are common to both the left and right ears, or even if the parameters themselves are different, some, such as noise suppression parameters, should be adjusted in a coordinated manner for both ears.

[0148] When such signal processing is included in the automatic adjustment process, the management of feedback data must be performed for both the left and right ears together. In this case, for example, as shown in the adjustment system 101 in Figure 19, the external cooperating device 40b may be configured such that the input audio buffer 71 and the feedback acquisition unit 72 are shared by the left ear hearing aid 50 and the right ear hearing aid 60.

[0149] Furthermore, all the functions provided by the external devices 40, 40a, and 40b may also be included on the hearing aid side. For example, the left ear hearing processing unit 20L and the right ear hearing processing unit 20R, which are examples of processing units, and the adjustment unit 10 may be mounted on the hearing aid side. Alternatively, the left ear hearing processing unit 20L and the right ear hearing processing unit 20R and the adjustment unit 10 may be mounted on a terminal device such as the external device 40 that outputs processed sound signal data to the hearing aid.

[0150] Furthermore, instead of storing all past data in the User Feedback DB74, recent data may be cached, and the main database may reside in the cloud. Also, the figures described above are merely examples and do not limit the location of each component related to this disclosure.

[0151] Furthermore, the effects described herein are merely illustrative and not limiting, and other effects may also occur.

[0152] Furthermore, this technology can also be configured as follows. (1) A processed sound generation step that generates a processed sound by acoustic processing using parameters that modify the sound collection function or hearing aid function of the sound output unit, An adjustment step in which the sound output unit is adjusted by parameters selected based on the parameters used in the sound processing and feedback to the processed sound output from the sound output unit. Information processing methods for information processing systems, including those mentioned above. (2) In the adjustment step described above, Based on the parameters used in the aforementioned sound processing and feedback on the processed sound output from the sound output unit, a method for selecting parameters suitable for the user is learned through machine learning, and the sound output unit is adjusted according to the parameters selected by the selection method. The information processing method described in (1) above. (3) In the adjustment step described above, The parameters used in the aforementioned acoustic processing and the feedback to the processed sound output from the sound output unit are acquired, and a prediction method is machine-learned to predict the feedback to the processed sound generated by acoustic processing using arbitrary parameters as a reward. Select the parameter that maximizes the predicted reward. The information processing method described in (2) above. (4) The sound output unit further includes a processed sound output step that outputs the processed sound. The information processing method described in any one of (1) to (3) above. (5) In the above-mentioned processed sound output step, The sound output unit outputs at least two or more processed sounds, each using different parameters for the sound processing. In the adjustment step described above, The parameters used in the acoustic processing of the two or more processed sounds and the feedback to the two or more processed sounds output from the sound output unit are acquired. The information processing method described in (4) above. (6) A display step that displays the speaker of the processed sound, A selection acceptance step that accepts an operation to select a preferred processing sound from the two or more processing sounds mentioned above. The information processing method described in (5) above, further comprising the above. (7) A display step that displays the speaker of the processed sound, A selection acceptance step that accepts a slider operation to select the degree of preference for the two or more types of processed sounds mentioned above. The information processing method described in (5) above, further comprising the above. (8) In the adjustment step described above, The results of manual parameter adjustments made by the user after listening to the outputted processed sound are obtained, and the method for selecting the parameters and predicting the reward are machine-learned based on the adjustment results. The information processing method described in (3) above. (9) In the adjustment step described above, Based on the parameters before and after manual adjustment by the user, and the user's predicted response to the processed sound using those parameters, the method for selecting the parameters and the method for predicting the reward are machine-learned. The information processing method described in (8) above. (10) In the adjustment step described above, Based on the user's feedback, to which a confidence level is added depending on whether the user's feedback is an actual response or a predicted response, the method for selecting the parameters and the method for predicting the reward are machine-learned. The information processing method described in (9) above. (11) In the adjustment step described above, The system estimates the user's situation upon hearing the outputted processed sound, and uses machine learning to determine the parameter selection method and the reward prediction method for each user's situation. The information processing method described in (3) above. (12) In the adjustment step described above, The system estimates the user's status from at least one of the following: information input by the user through operation or voice, the user's location information determined by GPS (Global Positioning System), the user's acceleration information detected by an acceleration sensor, and calendar information registered in an application program that manages the user's schedule. The information processing method described in (11) above. (13) In the adjustment step described above, The sound output section is adjusted according to parameters corresponding to the user's situation. The information processing method described in (11) or (12) above. (14) In the adjustment step described above, The parameters used in the aforementioned acoustic processing and feedback from multiple users who listened to the processed sound are obtained, and the method for selecting the parameters and the method for predicting the reward are learned through machine learning. The information processing method described in (3) above. (15) In the adjustment step described above, The parameters used in the aforementioned sound processing and the feedback from multiple users who listened to the processed sound are retrieved from a server that stores these parameters and the feedback from multiple users. The information processing method described in (14) above. (16) In the adjustment step described above, Based on the similarity to the user using the sound output unit to be adjusted, select a number of users from whom to obtain feedback. The information processing method described in (14) or (15) above. (17) In the adjustment step described above, Regarding the noise suppression parameters, the same parameters are selected for the right ear hearing aid and the left ear hearing aid. For parameters other than noise suppression, the parameters are selected individually for the right and left ear hearing aids. The information processing method described in any one of (1) to (16) above. (18) A processing unit that generates processed sound by acoustic processing using parameters that modify the sound collection function or hearing aid function of the sound output unit, An adjustment unit that adjusts the sound output unit using parameters selected based on the parameters used in the sound processing and feedback to the processed sound output from the sound output unit. An information processing system having (19) The system further includes a sound output unit that outputs the processed sound. The information processing system described in (18) above. (20) The aforementioned sound output section is, It is a hearing aid, The processing unit and the adjustment unit are, Mounted in the aforementioned hearing aid or a terminal device that outputs the processed sound signal data to the aforementioned hearing aid. The information processing system described in (18) or (19) above. [Explanation of Symbols]

[0153] 1. Information Processing System 10 Adjustment part 11 Agents 12. Reward Prediction Department 20 Processing Units 30 User Interface 40 External Connecting Devices 50. Hearing aid for the left ear. 60 Right ear hearing aid

Claims

1. A processed sound generation step that generates a processed sound by acoustic processing using parameters that modify the hearing aid function of the sound output unit, An adjustment step in which the sound output unit is adjusted by parameters selected based on the parameters used in the sound processing and feedback to the processed sound output from the sound output unit. Includes, In the adjustment step described above, Based on the parameters used in the aforementioned sound processing and feedback on the processed sound output from the sound output unit, a method for selecting parameters suitable for the user is learned through machine learning, and the sound output unit is adjusted according to the parameters selected by the selection method. In the adjustment step described above, The parameters used in the aforementioned acoustic processing and the feedback to the processed sound output from the sound output unit are acquired, and a prediction method is machine-learned to predict the feedback to the processed sound generated by acoustic processing using arbitrary parameters as a reward. Select the parameter that maximizes the predicted reward. Information processing methods for information processing systems.

2. The sound output unit further includes a processed sound output step that outputs the processed sound. The information processing method according to claim 1.

3. In the above-mentioned processed sound output step, The sound output unit outputs at least two or more processed sounds, each using different parameters for the sound processing. In the adjustment step described above, The parameters used in the acoustic processing of the two or more processed sounds and the feedback to the two or more processed sounds output from the sound output unit are acquired. The information processing method according to claim 2.

4. A display step that displays the speaker of the processed sound, A selection acceptance step that accepts an operation to select a preferred processing sound from the two or more processing sounds mentioned above. The information processing method according to claim 3, further comprising:

5. A display step that displays the speaker of the processed sound, A selection receiving step that accepts a slider operation to select the degree of preference for the two or more types of processed sounds mentioned above. The information processing method according to claim 3, further comprising:

6. In the adjustment step described above, The results of manual parameter adjustments made by the user after listening to the outputted processed sound are obtained, and the method for selecting the parameters and predicting the reward are machine-learned based on the adjustment results. The information processing method according to claim 1.

7. In the adjustment step described above, Based on the parameters before and after manual adjustment by the user, and the user's predicted response to the processed sound using those parameters, the method for selecting the parameters and the method for predicting the reward are machine-learned. The information processing method according to claim 6.

8. In the adjustment step described above, Based on the user's feedback, to which a confidence level is added depending on whether the user's feedback is an actual response or a predicted response, the method for selecting the parameters and the method for predicting the reward are machine-learned. The information processing method according to claim 7.

9. In the adjustment step described above, The system estimates the user's situation upon hearing the outputted processed sound, and uses machine learning to determine the parameter selection method and the reward prediction method for each user's situation. The information processing method according to claim 1.

10. In the adjustment step described above, The system estimates the user's status from at least one of the following: information input by the user through operation or voice, the user's location information determined by GPS (Global Positioning System), the user's acceleration information detected by an accelerometer, and calendar information registered in an application program that manages the user's schedule. The information processing method according to claim 9.

11. In the adjustment step described above, The sound output section is adjusted according to parameters corresponding to the user's situation. The information processing method according to claim 9.

12. In the adjustment step described above, The parameters used in the aforementioned acoustic processing and feedback from multiple users who listened to the processed sound are obtained, and the method for selecting the parameters and the method for predicting the reward are learned through machine learning. The information processing method according to claim 1.

13. In the adjustment step described above, The parameters used in the aforementioned sound processing and the feedback from multiple users who listened to the processed sound are retrieved from a server that stores these parameters and the feedback from multiple users. The information processing method according to claim 12.

14. In the adjustment step described above, Based on the similarity to the user using the sound output unit to be adjusted, select a number of users from whom to obtain feedback. The information processing method according to claim 12.

15. In the adjustment step described above, Regarding the noise suppression parameters, the same parameters are selected for the right ear hearing aid and the left ear hearing aid. For parameters other than noise suppression, the parameters are selected individually for the right and left ear hearing aids. The information processing method according to claim 1.

16. A processing unit that generates processed sound by acoustic processing using parameters that modify the hearing assistance function of the sound output unit, An adjustment unit that adjusts the sound output unit using parameters selected based on the parameters used in the sound processing and feedback to the processed sound output from the sound output unit. It has, The adjustment unit is, Based on the parameters used in the aforementioned sound processing and feedback on the processed sound output from the sound output unit, a method for selecting parameters suitable for the user is learned through machine learning, and the sound output unit is adjusted according to the parameters selected by the selection method. The adjustment unit is, The parameters used in the aforementioned acoustic processing and the feedback to the processed sound output from the sound output unit are acquired, and a prediction method is machine-learned to predict the feedback to the processed sound generated by acoustic processing using arbitrary parameters as a reward. Select the parameter that maximizes the predicted reward. Information processing system.

17. The system further includes a sound output unit that outputs the processed sound. The information processing system according to claim 16.

18. The aforementioned sound output section is, It is a hearing aid, The processing unit and the adjustment unit are, Mounted in the aforementioned hearing aid or a terminal device that outputs the processed sound signal data to the aforementioned hearing aid. The information processing system according to claim 17.

Images