Hearing aid fitting parameter configuration method based on speech recognition

Abstract

The application provides a hearing aid fitting parameter configuration method based on voice recognition. The method of the application inputs instructions in a voice mode through a voice acquisition and recognition module, replacing the traditional manual touch operation, so that the fitter can still adjust the parameters when the hands are occupied, avoiding the operation interruption caused by multi-task parallel. The double-mode intention matching module calls the fitter's closed vocabulary to isolate the user-level and fitter-level instructions, preventing the misuse of non-professionals; the fitting intention execution module routes the fitter intention identifier to the fitting page UI linkage module, executes the corresponding fitting operation in the target panel and generates the target fitting parameter, saving the fitter's manual operation steps of frequently switching between multiple function panels; the Bluetooth parameter issuing module issues the target fitting parameter to the hearing aid, completes the parameter update, so that the fitter can complete fine parameter adjustment without interrupting the current workflow, improving the fitting operation efficiency.

Description

Technical Field

[0001] This application relates to the field of hearing aid fitting technology, and in particular to a method for configuring hearing aid fitting parameters based on speech recognition. Background Technology

[0002] Hearing aid fitting is a specialized medical service that heavily relies on fine-tuning of parameters. During the fitting process, whether face-to-face or remotely, the audiologist needs to repeatedly adjust multiple parameters, such as gain curves, compression parameters, scene modes, noise cancellation switches, beamforming, and MPO (maximum power output), based on the hearing-impaired user's subjective feedback, to achieve the best hearing compensation effect. Traditional fitting methods primarily rely on sliders, buttons, tables, and drop-down menus on the fitting software interface, requiring the audiologist to continuously perform precise touch operations on the mobile terminal screen.

[0003] With the widespread adoption of smartphones and Bluetooth hearing aids, fitting parameters can be directly written to the hearing aid via Bluetooth from a mobile terminal, which improves the convenience of fitting to some extent. However, in actual fitting sessions, audiologists need to simultaneously observe the hearing-impaired user's reactions, record subjective feedback, and operate the interface on a mobile terminal. This multitasking significantly reduces operational efficiency and easily disrupts the communication rhythm with the user. Furthermore, detailed fitting pages typically contain multiple function panels, requiring frequent switching between panels to adjust different parameters, severely impacting the smoothness of the fitting process. Moreover, the voice control functions of existing hearing aid applications are usually only for end users (wearers or family members), providing basic operations such as volume adjustment and scene switching, without distinguishing user roles or designing independent voice control channels and dedicated command sets for the professional needs of audiologists. This prevents audiologists from completing fine parameter adjustments via voice, while ordinary users may cause parameter confusion due to accidentally triggering professional commands.

[0004] The aforementioned problems in the prior art result in low efficiency in hearing aid fitting operations. Therefore, how to improve the efficiency of fitting operations is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] This application provides a method for configuring hearing aid fitting parameters based on speech recognition, aiming to solve the technical problem of low efficiency in existing fitting operations.

[0006] This application provides a hearing aid fitting parameter configuration method based on speech recognition, applied to a hearing aid audiologist's voice digital assistant system. The hearing aid audiologist's voice digital assistant system includes a speech acquisition and recognition module, a dual-mode intent matching module, a fitting-specific vocabulary module, a fitting intent execution module, a fitting page UI linkage module, and a Bluetooth parameter distribution module. The hearing aid fitting parameter configuration method based on speech recognition includes the following steps: The voice input signal of the optician is acquired by the voice acquisition and recognition module, and the voice input signal is recognized to obtain the voice recognition text. Call the closed terminology library for audiologists that matches the audiologist mode in the dedicated terminology library module for audiologists; Based on the aforementioned optometrist mode, the dual-mode intent matching module performs text matching between the speech recognition text and the optometrist closed vocabulary to obtain the optometrist intent identifier. When the optometrist's intent identifier is a fitting-specific intent identifier, the fitting intent execution module routes the optometrist's intent identifier to the fitting page UI linkage module. The fitting operation corresponding to the audiologist's intent identifier is executed in the target panel through the fitting page UI linkage module, and the target fitting parameters are generated. The target fitting parameters are sent to the hearing aid via the Bluetooth parameter sending module to update the current fitting parameters of the hearing aid to the target fitting parameters.

[0007] This application provides a method for configuring hearing aid fitting parameters based on speech recognition. This method uses a speech acquisition and recognition module to input commands via voice, replacing traditional manual touch operation. This allows audiologists to adjust parameters even when their hands are occupied observing user reactions and recording feedback, avoiding operational interruptions caused by multitasking. A dual-mode intent matching module calls the audiologist's closed vocabulary, isolating user-level and audiologist-level commands. This ensures that the audiologist's dedicated intent identifier can only match fine-tuning commands, preventing misoperation by non-professionals. Furthermore, the fitting intent execution module routes the audiologist's intent identifier to the fitting page UI linkage module, directly executing the corresponding fitting operation and generating target fitting parameters in the target panel. This eliminates the manual steps of frequently switching between multiple function panels and searching for corresponding controls. Finally, the target fitting parameters are sent to the hearing aid via a Bluetooth parameter sending module, completing the parameter update. This allows audiologists to complete fine-tuning parameters without interrupting their current workflow, improving fitting efficiency. Attached Figure Description

[0008] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 A schematic diagram of the structure of an embodiment of a medical hygienist's voice digital assistant system provided in this application; Figure 2A flowchart illustrating an embodiment of a hearing aid fitting parameter configuration method based on speech recognition provided in this application; Figure 3 A schematic diagram of the basic adjustment interface in a digital voice assistant system for opticians provided in this application embodiment; Figure 4 A schematic diagram of a dynamic compression panel in a digital audio assistant system for opticians provided in this application; Figure 5 A schematic diagram of a microphone audio processing panel in a digital voice assistant system for opticians provided in this application; Figure 6 A schematic diagram illustrating the switching of different parameters in a dynamic compression interface of a digital audio assistant system for opticians, provided in this application; Figure 7 A schematic diagram illustrating the switching between different panels in a digital voice assistant system for opticians provided in this application; Figure 8 This is a schematic diagram illustrating the switching between different modes in a digital voice assistant system for opticians, as provided in this application.

[0010] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0011] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0012] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0013] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0014] This application provides a speech recognition-based method for configuring hearing aid fitting parameters, applied to a audiologist's voice digital assistant system. The audiologist's voice digital assistant system includes a speech acquisition and recognition module, a dual-mode intent matching module, a fitting-specific vocabulary module, a fitting intent execution module, a fitting page UI linkage module, and a Bluetooth parameter delivery module. The audiologist's voice digital assistant system also includes an audiologist's voice floating window module.

[0015] In a hearing aid fitting setting, audiologists typically need to perform multiple tasks simultaneously: observing the wearer's facial expressions and reactions to assess sound comfort, recording user feedback using a probe or recording board, and adjusting hearing aid parameters using a mobile application. Traditionally, audiologists must frequently look down at their phone screens, switching between multiple panels such as basic adjustments, dynamic compression, and audio processing, and clicking UI controls like sliders, buttons, and tables to fine-tune parameters. This method occupies the audiologist's hands, preventing them from simultaneously recording feedback or reassuring the wearer. Furthermore, the frequent interface switching disrupts the communication rhythm with the wearer, reducing fitting efficiency and user experience.

[0016] This embodiment provides a audiologist's voice digital assistant system deployed in a mobile terminal fitting application that communicates with hearing aids via Bluetooth. It is used to provide audiologists with a dedicated voice digital assistant to complete fine fitting parameter adjustments.

[0017] Please refer to Figure 1 , Figure 1 This is a schematic diagram of an embodiment of a digital voice assistant system for opticians provided in this application.

[0018] like Figure 1 As shown, the audiologist voice digital assistant system includes a voice acquisition and recognition module, a dual-mode intent matching module, a fitting-specific vocabulary module, a fitting intent execution module, a fitting page UI linkage module, a Bluetooth parameter distribution module, and an audiologist voice floating window module.

[0019] The voice acquisition and recognition module is used to acquire voice and output recognized text when explicitly triggered by the audiologist.

[0020] The dual-mode intent matching module includes a user mode and a audiologist mode. The audiologist mode is configured to allow matching of audiologist-specific intent identifiers and exclude user-side volume intent identifiers. The dual-mode intent matching module filters a unified instruction catalog through instruction category and audience fields, achieving dictionary isolation between the user mode and the audiologist mode.

[0021] The fitting-specific thesaurus module stores a set of closed instructions and keyword variations identified by the fitting-specific prefix.

[0022] The fitting intent execution module is used to route fitting-specific intents to the fitting page linkage interface.

[0023] The fitting page UI linkage module is used to call the same business methods as manual UI operations, updating fitting parameters, refreshing the interface, and triggering Bluetooth transmission. Before executing a fitting-specific intent, the fitting page UI linkage module automatically switches to the basic adjustment panel, dynamic compression panel, or audio processing panel corresponding to that intent. Please refer to... Figures 3 to 5 , Figure 3 A schematic diagram of a basic adjustment interface provided in this application, including a gain curve graph and a basic adjustment panel; Figure 4 A schematic diagram of a dynamic compression panel provided in this application, including a dynamic compression parameter adjustment interface; Figure 5 This is a schematic diagram of a microphone audio processing panel provided in this application, including an audio processing switch and a speed slider. The fitting page UI linkage module inherits the currently selected ear identifier and scene identifier, and only modifies the fitting parameter array corresponding to the ear identifier and scene. Please refer to... Figures 6 to 8 , Figure 6 A schematic diagram illustrating the switching of different parameters in a dynamic compression interface provided in this application; Figure 7 A schematic diagram illustrating the switching of different panels provided in this application; Figure 8 This is a schematic diagram illustrating the switching between different modes provided in this application.

[0024] The Bluetooth parameter delivery module is used to write the updated fitting parameters into the hearing aid.

[0025] The audio monitoring module is attached to the fitting adjustment page, providing voice on / off, press and hold to speak interaction, and feedback on the execution results. It can also be dragged and collapsed.

[0026] To enable those skilled in the art to gain a deeper understanding of the working principle of the audiologist's voice digital assistant system provided in this application, this application, in conjunction with actual hearing aid fitting scenarios, provides a detailed explanation of the functions and roles of each module in the aforementioned audiologist's voice digital assistant system: This system solves the problem of hands-free operation through a floating voice window module for audiologists. This module, a draggable and foldable floating window, is attached to the lower right corner of the fitting adjustment page. The audiologist simply needs to lightly touch the voice switch button on the floating window with one hand to activate the voice function, then press and hold the speaking area to speak the voice command. Releasing the hand completes the system's automatic recognition and execution. Throughout the process, the audiologist's other hand can always be used to record feedback or assist in operations, without needing to frequently look down at the screen. For example, when the audiologist observes that the wearer is uncomfortable with loud noise, they can simply say "reduce the volume," and the system will automatically adjust the corresponding parameters without interrupting the observation or recording.

[0027] To address the issue of numerous and hierarchical panels on the detailed fitting page, this system utilizes an automatic panel switching function within the fitting page UI linkage module. Before executing a fitting-specific intent, the UI linkage module automatically determines the panel category to which the intent belongs and calls the `voiceEnsureBasicMode()`, `voiceEnsureDynamicMode()`, or `voiceEnsureAudioMode()` methods to switch the current fitting adjustment page to the corresponding panel. For example, when the audiologist says "MPO increase," the system automatically switches from the current panel to the dynamic compression panel and performs the MPO increase operation; when the audiologist says "turn on noise cancellation," the system automatically switches to the audio processing panel and turns on the noise cancellation switch. The audiologist can smoothly execute voice commands between different panels without manually clicking panel switching buttons, maintaining the continuity of fitting communication.

[0028] To address the challenge of highly specialized hearing aid terminology being difficult for general voice assistants to understand accurately, this system employs a dedicated hearing aid terminology module and a dual-mode intent matching module in synergy. The dedicated terminology module constructs a closed set of instructions specific to the hearing aid fitting field, encompassing professional terms and their colloquial variations, including loudness curves (e.g., FIT_BIG_UP / DOWN), dynamic compression (e.g., FIT_MPO_UP / DOWN), sound quality processing (e.g., FIT_SOFTEN, FIT_SHARPEN), audio switching (e.g., FIT_DENOISE_ON / OFF), and scene modes (e.g., FIT_SCENE_OUTDOOR). The dual-mode intent matching module uses substring inclusion, keyword priority, confidence thresholds, and multi-rule conflict resolution strategies to stably map the audiologist's voice input to intent identifiers within the closed instruction set. For example, when a hygienist says "sharpen," it directly hits FIT_SHARPEN; when they say "MPO+," it hits FIT_MPO_UP. Unlike general voice assistants, it does not require open semantic understanding, thus ensuring the accuracy and controllability of intent recognition.

[0029] To address the risk of misoperation caused by the mixing of user-level and professional-level commands, this system employs a dual-mode intent matching module that separates user mode and audiologist mode. The dual-mode intent matching module filters a unified command catalog using command category and audience fields, achieving lexicon isolation between the two modes. User mode only allows terminal user-level commands such as volume, scene, and light sensing. audiologist mode inherits some user-level commands while additionally allowing the FIT_ series of fitting-specific commands, excluding the user-side VOL_ volume command to avoid semantic conflicts. All FIT_ prefix intents are only matchable in audiologist mode; ordinary users cannot access fitting-level parameter adjustment functions through the user mode's voice channel, fundamentally eliminating the risk of untrained personnel accidentally modifying fitting parameters.

[0030] Specifically, the dual-mode intent matching module includes a user mode and a audiologist mode. The audiologist mode is configured to allow matching of audiologist-specific intent identifiers and exclude user-side volume intent identifiers. The dual-mode intent matching module filters a unified instruction catalog through instruction category and audience fields to achieve lexicon isolation between the user mode and the audiologist mode.

[0031] User mode is for non-professional users such as wearers or their family members, and its entry point is the user voice control page. In this mode, the dual-mode intent matching module only allows end-user level commands, including basic functions such as volume adjustment, left / right ear switching, binaural synchronization, light sensor control, low battery reminder, and general interaction. Intent identifiers in user mode are prefixed with VOL_, such as VOL_UP, VOL_DOWN, VOL_LEFT, VOL_RIGHT, etc., and are used to control the daily wearing parameters of the hearing aid.

[0032] The user mode is designed to meet the daily operational needs of ordinary users while strictly limiting their access to fitting-level parameters. Ordinary users do not need to understand technical concepts such as MPO, dynamic compression, or gain curves; they can simply adjust volume and switch scenes using simple voice commands. Therefore, the user mode's vocabulary has been carefully selected, containing only commands relevant to daily use and excluding all fitting-specific intent identifiers with the FIT_ prefix.

[0033] The user-mode thesaurus isolation mechanism is implemented through the audience field. Each thesaurus entry in the unified instruction catalog contains an `audience` field, which identifies the applicable audience for the instruction. When filtering the available rule set, user mode only retains entries with `audience` set to `user` or `both`, completely excluding entries with `audience` set to `fitter`. For example, `VOL_UP` has `audience` set to `user`, while `FIT_BIG_UP` has `audience` set to `fitter`. When the system is in user mode, `VOL_UP` can be matched, but `FIT_BIG_UP` cannot.

[0034] User mode also includes scene-switching commands, such as SCENE_OUTDOOR and SCENE_QUIET, allowing ordinary users to switch between different usage scenarios. However, these scene-switching commands are executed directly via BLE writing, bypassing the fitting page UI linkage module, because ordinary users typically do not operate on the fitting adjustment page. User mode scene switching only updates the hearing aid's current working scenario and does not involve fine-tuning of fitting parameters.

[0035] The user mode's security level is end-user compliant, designed for ease of use and low accidental touch rate. Keywords in the dictionary use everyday colloquial expressions such as "speak louder," "switch to outdoor," and "left ear," avoiding technical jargon. The matching strategy is relatively lenient, with a lower confidence threshold to improve the success rate of voice interaction for ordinary users.

[0036] The audiologist mode is for professional audiologists, and its entry point is the fitting adjustment page floating window. In this mode, the dual-mode intent matching module opens fitting-specific intent identifiers and inherits some user-executable commands, but excludes the user-side VOL_ volume command to avoid semantic conflicts. Intent identifiers in audiologist mode are prefixed with FIT_, such as FIT_BIG_UP, FIT_MPO_DOWN, FIT_SHARPEN, FIT_DENOISE_ON, etc., and are used to complete fine-tuning of fitting parameters.

[0037] The Audiologist Mode is designed to provide professional audiologists with an efficient, accurate, and secure voice control channel. Audiologists frequently need to adjust multiple parameters, such as gain curves, compression parameters, sound quality processing, and noise reduction switches, during on-site fittings. Traditional manual operation is inefficient. Audiologist Mode, through a closed dictionary and precise matching, enables audiologists to quickly drive parameter adjustments using short natural language phrases, while ensuring the accuracy and security of commands.

[0038] The audiologist mode's dictionary isolation mechanism is also implemented through the audience field, but the filtering rules are the opposite of the user mode. When filtering the available rule set, the audiologist mode retains entries with "audience" set to "fitter" or "both," while excluding entries with "audience" set to "user" and belonging to the volume category. Specifically, all FIT_ series commands with "audience" set to "fitter" are enabled; some commands with "audience" set to "both" (such as scene switching) are also enabled; and VOL_ series commands with "audience" set to "user" are excluded to avoid semantic conflicts with the dedicated volume sliders (FIT_VOL_UP / DOWN) on the fitting page.

[0039] The closed lexicon in the audiologist mode is finely divided according to the dimensions of audiologist services, including the following categories: loudness curves, dynamic compression, audio switches, scene modes, and audiologist volume. The audiologist mode maintains a professional-grade level of security isolation, designed with precision, controllability, and traceability in mind. All FIT_ prefix intents are only matchable in audiologist mode and are inaccessible to ordinary users. Keywords in the lexicon use audiologist terminology and its colloquial variations, such as "MPO," "transient noise," and "feedback suppression." The matching strategy is relatively strict, with a high confidence threshold (e.g., 0.7) to ensure accurate intent recognition. No BLE write operations are performed on unmatched instructions to avoid accidental operations.

[0040] User mode and audiologist mode work collaboratively through a dual-mode intent matching module. This module also handles mode switching and state management. The system defines two modes through the VoiceControlMode enumeration and maintains the current mode state throughout the application lifecycle. Mode switching is triggered by: automatically switching to user mode when the user navigates to the user voice control page; automatically switching to audiologist mode when the audiologist enters the fitting adjustment page and opens the voice pop-up; and automatically releasing the current mode when the user or audiologist exits the corresponding page. When a user enters from the user voice control page, the system activates user mode, and the dual-mode intent matching module loads user mode filtering rules, matching only user-level commands. When an audiologist triggers voice commands from the fitting adjustment page pop-up, the system activates audiologist mode, and the dual-mode intent matching module loads audiologist mode filtering rules, enabling fitting-specific commands. The two modes switch mutually exclusively, ensuring that only one mode is active at any given time, avoiding mode confusion and command conflicts.

[0041] To address the issue of asynchronous operation between the interface, memory, and Bluetooth caused by the inconsistency between the voice execution path and the manual operation path, this system employs a UI-based execution mechanism that links the fitting intent execution module and the fitting page UI linkage module. The fitting intent execution module routes the fitting-specific intent to the fitting page UI linkage module, which then calls the exact same business method chain as the manual UI operation to execute the fitting operation. For example, when the audiologist says "sharpen," the system calls `voiceApplySoundQuality(sharpen:true)`, which is identical to the method called when manually clicking the "sharpen" button. Both methods select the corresponding arrays based on `currentEarType` and `sceneType`, call `HearingAdjustTools.getSharpenData` to calculate the new curve, refresh the chart using `changeSceneData`, and finally asynchronously write the data to the hearing aid using `clickWrite()`. Similarly, when the audiologist says "Switch to outdoor mode," the system calls applySceneSelection(type:2), the same method used when manually selecting outdoor mode. This synchronously updates the scene button title, gain curve, dynamic compression table, and audio header parameters, and writes the information to the scene configuration via the Bluetooth API. This UI-path-based execution mechanism ensures that the interface display after voice triggering, the in-memory fitting data, and the Bluetooth transmission remain consistent, avoiding inconsistencies such as "the interface changes but the device hasn't written the information" or "the device has written the information but the interface hasn't refreshed."

[0042] Furthermore, the audiologist voice digital assistant system provided in this application further improves fitting efficiency by inheriting the current ear type and scene context. When performing fitting operations, the UI linkage module on the fitting page reads the values ​​of `currentEarType` and `sceneType`, modifying only the fitting parameter arrays for the corresponding ear type and scene. For example, when the system currently selects the left ear and the outdoor scene, the audiologist saying "loudness increase" only modifies the loudness gain curve for the left ear in the outdoor scene, without affecting the parameters for the right ear or other scenes. The audiologist does not need to manually select the ear type and scene before each instruction to perform parameter adjustments in the correct context, making it particularly suitable for binaural fitting and multi-scene comparison and adjustment scenarios.

[0043] The audiologist voice digital assistant system provided in this application effectively solves existing technical problems such as audiologists' hands being occupied, frequent interface switching, difficulty in understanding professional terminology, high risk of misoperation, and inconsistent parameter synchronization through the collaborative work of various modules, and significantly improves the efficiency, safety and reliability of hearing aid fitting.

[0044] Please refer to Figure 2 , Figure 2 This is a flowchart illustrating an embodiment of a hearing aid fitting parameter configuration method based on speech recognition provided in this application.

[0045] like Figure 2 As shown, the speech recognition-based hearing aid fitting parameter configuration method is applied to the aforementioned hearing aid digital assistant system for audiologists, and the method includes steps S101 to S106.

[0046] S101. The voice input signal of the audiologist is acquired through the voice acquisition and recognition module, and the voice input signal is recognized to obtain the voice recognition text.

[0047] In hearing aid fitting scenarios, audiologists often face the challenge of having their hands occupied. During face-to-face or remote fittings, audiologists need to simultaneously observe the wearer's facial expressions and reactions to assess sound comfort, record the user's subjective feedback using a probe or recording board, and adjust hearing aid parameters using a mobile application. Traditional methods rely primarily on sliders, buttons, tables, and drop-down menus on the fitting software interface. audiologists must frequently look down at their phone screens, switching between multiple panels such as basic adjustments, dynamic compression, and audio processing, and clicking various UI controls to fine-tune parameters. This method keeps the audiologist's hands occupied, preventing them from simultaneously recording feedback or reassuring the wearer. Furthermore, frequent interface switching disrupts the communication rhythm with the wearer, reducing fitting efficiency and user experience. In addition, existing smart speaker-based solutions employ continuous listening and open semantic understanding, making intent uncontrollable and unsuitable for medical / fitting scenarios; fully offline recognition solutions have large model sizes and limited support for the elderly and specialized terminology; and simple "speech-to-text + cloud semantic API" solutions cannot guarantee command whitelist security.

[0048] To address the aforementioned technical challenges, this step utilizes a voice acquisition and recognition module. Under explicit triggering by the audiologist, the module acquires voice data and outputs recognized text, enabling the audiologist to adjust parameters using short natural language sentences without frequent touchscreen operations.

[0049] Specifically, the voice acquisition and recognition module's acquisition and recognition process is triggered by the audiologist's voice floating window module. This module, implemented as a SpeechFittingVoiceOverlay, is attached to the fitting adjustment page as a draggable, collapsible floating window. The floating window includes interactive elements such as a voice on / off button, a press-to-speak area, and a status bar. Microphone and voice recognition permissions are only requested after the audiologist actively turns on the floating window, avoiding the problem of excessive permission requests in non-fitting scenarios.

[0050] The audiologist's voice floating window module, attached to the fitting adjustment page, receives the audiologist's voice switch trigger operation and press-to-talk operation. After the voice input signal is collected, the audiologist's voice floating window module provides feedback on the execution result to the audiologist.

[0051] Specifically, after the audiologist clicks the voice switch button, a floating window enters the data collection state, and the status bar displays the "Press and hold to speak" prompt. The audiologist presses their finger on the speaking area, and the voice acquisition and recognition module begins to collect the voice input signal. This module uses a non-continuous monitoring method of pressing and holding to speak, collecting audio only while the audiologist is pressing and holding, and stopping immediately after releasing the finger, avoiding false triggering and privacy risks caused by continuous monitoring. During the collection process, the voice acquisition and recognition module captures analog audio signals through the mobile terminal's microphone hardware, and obtains a digital audio stream after analog-to-digital conversion.

[0052] The voice acquisition and recognition module preprocesses the digital audio stream before sending it to the voice recognition engine. The preprocessing includes steps such as noise reduction, echo cancellation, and automatic gain control to improve recognition accuracy in the complex acoustic environment of the fitting site.

[0053] For example, a speech recognition engine can adopt a hybrid architecture that combines online speech recognition with local auxiliary recognition: the online recognition engine is deployed on a cloud server and has powerful acoustic and language models to support accurate recognition of professional terminology; the local auxiliary recognition engine serves as a fallback solution and is activated when the network is unavailable or the latency is too high to ensure the continuity of speech functions.

[0054] The speech recognition engine outputs recognition results, including partial and final results. The partial result is a real-time intermediate result, continuously returned while the audiologist is speaking, used for real-time feedback display in the floating status bar, such as "Recognizing: Loud...". The final result is the final confirmation result, returned after the audiologist releases their finger, and serves as the input text for subsequent intent matching. The speech acquisition and recognition module outputs the final result as a string, i.e., the speech-recognized text.

[0055] S102. Call the closed terminology library for audiologists that matches the audiologist mode in the dedicated terminology library module for audiologists.

[0056] In hearing aid fitting scenarios, fitting terminology differs significantly from everyday speech, making it difficult for general voice assistants to accurately understand. During face-to-face or remote fitting, audiologists need to repeatedly adjust multiple parameters such as gain curves, compression parameters, scene modes, noise cancellation switches, beamforming, and MPO (maximum power output) based on user feedback. These parameters involve numerous technical terms, such as "loud," "medium," "MPO," "transient noise," and "feedback suppression," which differ significantly from everyday speech. Current technologies, such as smart speaker solutions using continuous listening and open semantic understanding, lack uncontrollable intent and are unsuitable for medical / fitting scenarios; fully offline recognition solutions have large model sizes and limited support for the elderly and technical terminology; simple "speech-to-text + cloud semantic API" solutions cannot guarantee command whitelist security and cannot stably map speech recognition results to the closed command set of the fitting domain. Mixing fitting-level commands with user-level commands in the same open semantic channel could lead to untrained personnel accidentally modifying fitting parameters, resulting in a high risk of misoperation.

[0057] To address the aforementioned technical challenges, this step utilizes a dedicated fitting terminology module to store audiologist-specific instructions and keyword variations, constructing a closed and controllable instruction set. This ensures that speech recognition results can be stably mapped to a closed instruction set within the fitting field.

[0058] Specifically, the fitting-specific terminology module stores a set of closed instructions and keyword variations identified by a fitting-specific prefix. The fitting specialist terminology is finely divided according to the dimensions of fitting business, covering categories such as loudness curves, speech curves, dynamic compression, sound quality processing, audio switches, scene modes, and fitting volume.

[0059] The loudness curve category includes FIT_BIG_UP and FIT_BIG_DOWN (increase / decrease loud volume curves), FIT_MID_UP and FIT_MID_DOWN (increase / decrease mid-range volume curves), and FIT_SMALL_UP and FIT_SMALL_DOWN (increase / decrease loud volume curves). These commands correspond to the gain curve adjustment function in the basic adjustment panel. Please refer to... Figure 3 , Figure 3 The left image shows the loudness adjustment area of ​​the basic adjustment panel, containing increase and decrease buttons for four dimensions: loud, medium, soft, and speech. Typical speech commands include "loud increase," "loud decrease," "medium increase," and "soft decrease." When the audiologist observes discomfort from a particular loudness level, they can directly adjust the gain value of the corresponding loudness curve using these commands, both via system invocation and manual click. Figure 3 The left figure shows the same business method chain when corresponding to the add and remove buttons, updating the gain curve and refreshing the chart.

[0060] The speech curve category includes FIT_LANG_UP and FIT_LANG_DOWN (increase / decrease speech curve). These commands are specifically designed to adjust the gain curve in the speech frequency range to optimize the wearer's perceived clarity in speech communication scenarios. Typical speech expressions include "speech increase" and "speech decrease".

[0061] The dynamic compression category includes FIT_MPO_UP and FIT_MPO_DOWN (increase / decrease MPO, i.e., maximum power output), and FIT_GEAR_UP and FIT_GEAR_DOWN (increase / decrease gear level). These commands correspond to the parameter adjustment functions in the dynamic compression panel. Please refer to... Figure 4 , Figure 4 The image shows the dynamic compression panel, displaying the MPO values ​​and adjustment buttons for each frequency point in a table format, including "MPO Increase," "MPO Decrease," "Level Increase," and "Level Decrease." The MPO parameter controls the maximum sound output level of the hearing aid, preventing excessive sound from causing discomfort; the level parameter controls the compression ratio, affecting the dynamic range of the sound. When the audiologist says "MPO Increase," the system can activate the adjustment, or it can be manually clicked. Figure 4 The same business method chain is used when the MPO adds a button, updating the MPO parameters and refreshing the dynamic compression table.

[0062] The audio processing category includes FIT_SOFTEN and FIT_SHARPEN. These commands correspond to the audio processing functions in the basic adjustment panel. Please refer to [link / reference]. Figure 3 , Figure 3 The image on the right shows the sound quality area of ​​the basic adjustment panel, which includes two buttons: "Soften" and "Sharpen." When the audiologist says "Soften," the system activates the sound quality, and manual clicking is also possible. Figure 3 The "Soften" button shown in the right image uses the same business method chain to make the sound softer and more comfortable, suitable for wearers sensitive to sharp sounds; when the audiologist says "Sharpen," the system calls and manually clicks the button. Figure 3 The "Sharpen" button shown in the right figure uses the same business method chain to enhance the detail and clarity of the sound, making it suitable for scenarios that require higher speech resolution.

[0063] The audio switch category includes FIT_BEAM_ON and FIT_BEAM_OFF (beamforming on / off), FIT_ORIGINAL_ON and FIT_ORIGINAL_OFF (original sound processing on / off), FIT_DENOISE_ON and FIT_DENOISE_OFF (noise reduction on / off), FIT_WIND_ON and FIT_WIND_OFF (wind noise suppression on / off), FIT_TRANSIENT_ON and FIT_TRANSIENT_OFF (transient noise suppression on / off), and FIT_SQUEAL_ON and FIT_SQUEAL_OFF (feedback suppression on / off). These commands correspond to various function switches in the audio processing panel; please refer to [link / reference]. Figure 5 , Figure 5 The microphone audio processing panel shown includes multiple switches and sliders for beamforming, feedback suppression, WDRC, noise cancellation, wind noise reduction, transient noise reduction, and original sound processing. Typical voice commands include "Turn on noise cancellation," "Turn off wind noise," and "Turn on feedback suppression." The audiologist can flexibly enable or disable various audio processing functions based on the wearer's environment and subjective feedback, through both system calls and manual clicks. Figure 5 The corresponding switch uses the same business method chain to update the switch status and refresh the panel display.

[0064] The scene mode category includes FIT_SCENE_GENERAL (General Mode), FIT_SCENE_OUTDOOR (Outdoor Mode), FIT_SCENE_NOISY (Noisy Mode), FIT_SCENE_QUIET (Quiet Mode), FIT_SCENE_TV (TV Mode), and FIT_SCENE_CONVERSATION (Conversation Mode). These commands are used to switch the hearing aid's operating environment; please refer to [link / reference]. Figure 8 , Figure 8The drop-down menu for mode switching is shown, containing six scene modes: General Mode, Outdoor Mode, Noisy Mode, Quiet Mode, TV Mode, and Conversation Mode. Typical voice prompts include "Switch to Outdoor Mode" and "Switch to Quiet Mode." When the audiologist says "Switch to Outdoor Mode," the system will invoke the mode, and manual clicking will also trigger the mode. Figure 8 The “Outdoor Mode” option shown uses the same applySceneSelection method. In different scene modes, the hearing aid’s gain curve, compression parameters, noise reduction strategy and other configurations are different to adapt to different acoustic environments.

[0065] The volume control categories for fitting pages include FIT_VOL_UP and FIT_VOL_DOWN (volume slider adjustment on the fitting page). These commands are specifically designed to control the volume slider on the fitting page, and are distinct from the VOL_UP and VOL_DOWN commands in user mode to avoid semantic conflicts. Typical voice expressions include "turn up the volume" and "turn down the volume."

[0066] Each dictionary entry includes fields such as intentId, main keyword, variant keywords, category, and audience. The intentId uses the FIT_ prefix naming convention to ensure clear distinction from the VOL_ prefix intents in user mode. The main keyword is the most commonly used speech expression, while variant keywords are synonyms or near-synonyms to improve speech matching coverage and success rate. The category field is used for subsequent automatic panel switching, and the audience field is used for pattern filtering in the dual-mode intent matching module.

[0067] The fitting dictionary is maintained using a dual-channel approach, employing both JSON and Swift build rules to ensure its maintainability and scalability. The product-side Markdown dictionary is exported to JSON format via the `export_voice_intents.py` script for easy editing and management by product personnel. The fitting specialist's FIT rules are injected at compile time using the `VoiceFittingLexicon.buildRules()` method, ensuring the stability of core instructions. At runtime, `VoiceIntentCatalog` merges the JSON dictionary and Swift build rules into a unified instruction directory, filters by pattern, and outputs it to the dual-pattern intent matching module. This dual-channel maintenance mechanism ensures both the dynamic updating capability of the dictionary and the reliability of core instructions.

[0068] This embodiment constructs a closed lexicon for fitting terminology such as MPO, transient noise, and howling suppression, solving the problem that general voice assistants have difficulty accurately understanding professional terms. The closed lexicon ensures the security of the command whitelist and avoids the risk of non-professionals accidentally modifying fitting parameters. The lexicon structure, which is finely divided according to the dimensions of fitting business, makes it easy for fitters to quickly locate and use the corresponding commands according to the adjustment goals. The dual-channel maintenance mechanism takes into account both the dynamic updating of the lexicon and the stability of core commands, improving the maintainability of the system.

[0069] S103. Based on the audiologist mode, the dual-mode intent matching module performs text matching between the speech recognition text and the audiologist closed vocabulary to obtain the audiologist intent identifier.

[0070] In hearing aid fitting scenarios, fitting terminology differs significantly from ordinary user speech, making it difficult to reliably map speech recognition results to the closed instruction set of the fitting domain. Existing technologies, general-purpose voice assistants employ open semantic understanding, resulting in uncontrollable intent and failing to guarantee accurate mapping of recognition results to fitting parameter adjustment functions. Simple keyword matching schemes lack confidence assessment and conflict resolution mechanisms, easily leading to mismatches in scenarios with polysemous words or similar instructions. Furthermore, mixing fitting-level and user-level instructions in the same open semantic channel can lead to untrained personnel mistakenly modifying fitting parameters, posing a high risk of operational errors. Therefore, there is an urgent need for an intent matching scheme tailored to the role of hearing aid audiologists, with closed and controllable instructions, and stable and reliable matching results.

[0071] To address the aforementioned technical challenges, this step employs a dual-mode intent matching module to filter the vocabulary by role mode and output intent identifiers and confidence scores, thereby achieving a stable mapping of speech recognition results to a closed set of instructions in the fitting domain.

[0072] Specifically, the dual-mode intent matching module filters the unified instruction catalog through instruction category and audience fields, achieving lexicon isolation between the user mode and the fitter mode. The dual-mode intent matching module provides two mutually exclusive voice control modes: user mode and fitter mode.

[0073] User mode is designed for non-professional users, such as wearers or their family members, and is accessed through the user voice control page. In this mode, the dual-mode intent matching module only allows end-user-level commands, including basic functions such as volume adjustment, left / right ear switching, binaural synchronization, light sensor control, low battery reminder, and general interaction. Intent identifiers in user mode are prefixed with "VOL_", and keywords in the dictionary use everyday colloquial expressions, such as "louder," "switch to outdoor," and "left ear." When filtering the available rule set, user mode only retains entries with "audience" set to "user" or "both," completely excluding entries with "audience" set to "fitter."

[0074] The fitter mode is applicable to professional fitters, and its entry is a floating window on the fitting adjustment page. In this mode, the dual-mode intent matching module opens dedicated fitter intent identifiers, while inheriting part of the instructions executable by the user, but excluding the user-side VOL_ volume instruction to avoid semantic conflicts. The intent identifiers in the fitter mode are prefixed with FIT_, and the keywords adopt fitting professional terms and their colloquial variants. When the fitter mode filters the available rule set, entries whose audience is fitter or both are retained, while entries whose audience is only user and belong to the volume category are excluded. All FIT_-prefixed intents can only be matched in the fitter mode, and ordinary users cannot trigger fitter-level parameter adjustment functions through the voice channel in the user mode.

[0075] Further, the step of performing text matching on the speech recognition text and the fitter closed lexicon through the dual-mode intent matching module based on the fitter mode to obtain a fitter intent identifier comprises: performing normalization processing on the speech recognition text to obtain a normalized text; filtering the fitter closed lexicon based on the fitter mode to obtain an available rule set; matching each rule in the available rule set with the normalized text based on a preset matching strategy, and calculating the matching confidence of each rule; and outputting the corresponding fitter intent identifier when the highest confidence is greater than or equal to the preset confidence threshold.

[0076] In an embodiment, the performing normalization processing on the speech recognition text may include, but is not limited to: converting to lowercase to unify the case form of the text; removing spaces and line breaks to eliminate format interference; removing Chinese and English punctuation marks to retain only plain text content; removing modal particles, such as meaningless words without actual semantics like "um", "ah", "that", "please", "help me", "a moment", etc.; optional processing of erhua suffixes, for example, normalizing "dà shēngr" (loud with erhua) to "dà shēng" (loud). The normalization processing eliminates noise and redundant information in the speech recognition text, and improves the accuracy of subsequent matching.

[0077] The dual-mode intent matching module reads the current mode state, and loads the corresponding filtered rule set according to the state. When the current mode is the fitter mode, entries whose audience is fitter or both are retained, and VOL_ series instructions whose audience is only user are excluded, so as to obtain the available rule set under the fitter mode.

[0078] Each rule in the available rule set is matched against the normalized text. The matching strategy can employ substring inclusion detection, which checks whether the normalized text contains the main keyword or variant keyword from the rule. The calculation of matching confidence can follow these principles: the matching weight of the main keyword is higher than that of variant keywords, for example, the main keyword "loud plus" has a weight of 1.0, while the variant keyword "increase the volume" has a weight of 0.8; weighting for complete equality, if the normalized text is completely equal to the keyword, additional weight is added on top of the base weight; and priority is given to longer keywords, in the case of the same confidence, the rule with the longer keyword is selected because longer keywords usually have higher discriminative power.

[0079] The preset confidence level can be set according to actual application needs, such as 0.7. If the highest confidence level is lower than this threshold, it is considered a no-match, no BLE write operation is performed, and the status bar will display "Unrecognized, please try again". When multiple rules conflict, the one with the highest confidence level is selected. If the scores are the same, the longer keyword is selected to ensure that the most accurate intent is selected when multiple rules match simultaneously. For example, if the speech recognition text is "loud plus", and it matches both FIT_BIG_UP (confidence level 0.95, keyword "loud plus") and FIT_LOUDNESS_UP (confidence level 0.85, keyword "loudness"), the system will select FIT_BIG_UP with the higher confidence level for execution.

[0080] This embodiment fundamentally reduces the probability of misoperation by isolating the user mode and the audiologist mode vocabulary, and by not opening the fitting instructions to ordinary users. Through substring inclusion, keyword priority, confidence threshold, and multi-rule conflict resolution strategies, it achieves a stable mapping from speech recognition results to a closed set of instructions in the fitting domain, solving the problem of uncontrollable intent in general voice assistants. Normalization processing eliminates noise interference in the speech recognition text, further improving matching accuracy. Strict confidence threshold and non-match protection mechanism avoid parameter misadjustment caused by low-confidence recognition results.

[0081] In one embodiment, the audiologist voice digital assistant system can employ two execution modes: confirmation execution mode and instant execution mode.

[0082] In one embodiment, when the audiologist's voice digital assistant system is configured in confirmation execution mode, the voice acquisition and recognition module acquires the audiologist's continuous voice input signal and performs voice recognition on the voice input signal to obtain voice-recognized text containing at least one voice command; the voice-recognized text is displayed through the audiologist's voice floating window module, and the audiologist's confirmation operation on the voice-recognized text is received; upon receiving the confirmation operation, the dual-mode intent matching module performs text matching between the voice-recognized text and the audiologist's closed vocabulary to obtain the audiologist's intent identifier.

[0083] In the confirmation execution mode, the voice acquisition and recognition module acquires the continuous voice input signal of the audiologist and performs voice recognition on the voice input signal to obtain voice-recognized text containing at least one voice command. The voice-recognized text is displayed through the audiologist's voice floating window module, and the audiologist's confirmation operation on the voice-recognized text is received. Upon receiving the confirmation operation, the dual-mode intent matching module performs text matching between the voice-recognized text and the audiologist's closed vocabulary to obtain the audiologist's intent identifier.

[0084] The confirmation execution mode is suitable for scenarios requiring high-precision recognition, such as initial fitting, significant parameter adjustments, or environments with high noise levels. In this mode, the system adds a manual confirmation step, allowing the audiologist to confirm the accuracy of the voice recognition results in the floating status bar before execution, thus avoiding incorrect parameter adjustments due to recognition errors.

[0085] Specifically, the audiologist first triggers voice input through a floating voice window module attached to the fitting adjustment page. The audiologist clicks the matching button on the floating voice window, and the system enters voice acquisition mode. Unlike the press-to-talk mode in immediate execution mode, in confirmed execution mode, the audiologist does not need to hold down the button but directly speaks the voice command. The voice acquisition and recognition module collects the audiologist's continuous voice input signal, which may contain one or more voice commands. For example, the audiologist can continuously say "Turn on noise cancellation, turn up the volume," and the system will collect this as a continuous voice input.

[0086] The speech acquisition and recognition module performs speech recognition on the acquired speech input signal. Preprocessing includes noise reduction, echo cancellation, and automatic gain control to improve recognition accuracy. The speech recognition engine adopts a hybrid architecture combining online speech recognition and local assisted recognition to output the recognition result. Since the confirmation execution mode supports continuous speech input, the speech recognition engine needs to segment the continuous speech into sentences to identify at least one speech command contained within it, ultimately obtaining the speech recognition text containing at least one speech command.

[0087] After obtaining the speech-recognized text, the system does not directly perform matching and operations. Instead, it displays the speech-recognized text through a floating window module for the audiologist. The floating window status bar displays the recognition result in full text form, for example, "Recognition result: Turn on noise cancellation, increase volume." The audiologist reviews the recognized text on the floating window interface to verify whether it matches the instruction they intended. If the recognition result is inaccurate, such as misrecognizing "turn on noise cancellation" as "turn on noise reduction," the audiologist can choose to re-enter the text or ignore the result; the system will not perform any operation.

[0088] After confirming that the system has correctly recognized the voice content, the audiologist sends a confirmation request to the system by clicking the confirmation button on the voice pop-up window. This confirmation button is an independent interactive element on the pop-up window interface and is only displayed in confirmation execution mode. After the audiologist clicks the confirmation button, the system receives the confirmation signal and triggers the subsequent intent matching process.

[0089] Upon receiving the confirmation operation, the dual-mode intent matching module performs text matching between the speech recognition text and the audiologist's closed vocabulary. The dual-mode intent matching module first normalizes the speech recognition text, including converting it to lowercase, removing spaces and line breaks, removing Chinese and English punctuation, and removing interjections. Then, based on the current audiologist mode, it filters the audiologist's closed vocabulary to obtain a set of available rules. Next, based on a preset matching strategy, it performs substring inclusion detection on each rule in the available rule set and the normalized text, calculating the matching confidence of each rule. When the highest confidence is greater than or equal to a preset confidence threshold, the corresponding audiologist intent identifier is output. If the speech recognition text contains multiple instructions, such as "turn on noise reduction, increase volume," the dual-mode intent matching module matches them one by one, obtaining two intent identifiers: FIT_DENOISE_ON and FIT_VOL_UP.

[0090] This embodiment significantly reduces the risk of parameter misadjustment due to voice recognition errors by adding a manual confirmation step, making it particularly suitable for critical scenarios such as first-time fitting or significant parameter adjustments; it supports continuous voice input, allowing audiologists to speak multiple commands at once, and the system displays the complete recognition results before unified confirmation and execution, improving the efficiency of batch operations; audiologists can cancel or re-enter at any time before confirmation, enhancing the controllability and security of the operation.

[0091] In another embodiment, when the audiologist's voice digital assistant system is configured in instant execution mode, the audiologist's press-to-talk operation is detected by the audiologist's voice floating window module; during the press-to-talk operation, the audiologist's voice input signal is acquired in real time by the voice acquisition and recognition module, and the voice input signal is recognized to obtain real-time recognized text; after the press-to-talk operation is detected to be released, the real-time recognized text and the audiologist's closed vocabulary are matched by the dual-mode intent matching module to obtain the audiologist's intent identifier.

[0092] In real-time execution mode, the optician's press-to-talk operation is detected via the optician's voice floating window module. During the press-to-talk operation, the voice acquisition and recognition module acquires the optician's voice input signal in real time and performs voice recognition on the voice input signal to obtain real-time recognized text. After the press-to-talk operation is detected to be released, the dual-mode intent matching module performs text matching between the real-time recognized text and the optician's closed vocabulary to obtain the optician's intent identifier.

[0093] The instant execution mode is suitable for scenarios requiring rapid response, such as when the audiologist is familiar with voice interaction, the ambient noise is low, or frequent fine-tuning of parameters is needed. Execution occurs immediately upon releasing the button, without additional confirmation, thus improving operational efficiency. In this mode, the system eliminates the manual confirmation step; the recognition result is executed immediately after the audiologist releases the button, resulting in fast response and smooth operation.

[0094] Specifically, the audiologist triggers voice input via a floating voice window module attached to the fitting and adjustment page. The audiologist clicks the matching button on the floating voice window and presses and holds the speaking area. Upon detecting this press-and-hold action, the system immediately enters real-time audio recording mode. Unlike the click-triggered confirmation execution mode, the real-time execution mode uses a press-and-hold interaction, recording audio only while the finger is held down. Recording stops immediately upon releasing the finger, avoiding accidental triggering and privacy risks caused by continuous monitoring.

[0095] During the press-to-talk operation, the voice input signal of the audiologist is acquired in real time by the voice acquisition and recognition module. The voice acquisition and recognition module captures analog audio signals through the microphone hardware of the mobile terminal and obtains a digital audio stream after analog-to-digital conversion. At the same time, the digital audio stream is preprocessed, including noise reduction, echo cancellation, and automatic gain control, to improve the recognition accuracy in the complex acoustic environment of the fitting site.

[0096] The speech acquisition and recognition module performs real-time speech recognition on the preprocessed audio stream to obtain real-time recognized text. The speech recognition engine continuously returns partial recognition results, i.e., real-time intermediate results, while the speaker is pressed and held. These partial recognition results are used for real-time feedback display in the floating status bar, such as "Recognizing: Louder..." or "Recognizing: Open...", providing audiologists with progress feedback so they understand what the system is recognizing. However, partial recognition results are only intermediate states, unstable, and subject to change; the system does not perform intent matching or execution on these results.

[0097] After detecting that the press-to-talk operation has been released, the voice acquisition and recognition module returns a final recognition result, which is the final confirmation result. The final recognition result is the final recognized text of the entire speech when the audiologist releases the button, and its stability is higher than that of the partial recognition result. The voice acquisition and recognition module outputs the final recognition result as a string, which serves as the input for subsequent intent matching.

[0098] After the press-to-talk operation is detected as released, the dual-mode intent matching module performs text matching between the real-time recognized text and the audiologist's closed vocabulary. The real-time recognized text here is the final recognition result. The dual-mode intent matching module first performs normalization processing on the final recognized text, including converting to lowercase, removing spaces and line breaks, removing Chinese and English punctuation, and removing interjections. Then, based on the current audiologist mode, it filters the audiologist's closed vocabulary to obtain a usable rule set. Next, based on a preset matching strategy, it performs substring inclusion detection on each rule in the usable rule set and the normalized text, calculating the matching confidence of each rule. When the highest confidence is greater than or equal to a preset confidence threshold, the corresponding audiologist intent identifier is output.

[0099] In addition, the instant execution mode includes a deduplication mechanism. The system maintains a recently executed record table, recording the intentId and execution time after each execution intent. When a new matching result is generated, the system checks the recently executed record table. If the intentId of the currently matched intentId is the same as that of the most recently executed intentId and the time interval is less than a threshold (e.g., 2 seconds), it is determined to be a duplicate execution, and the current execution is skipped. The system then sends a "Deduplication Completed" message to the floating status bar, preventing the audiologist from performing repeated operations due to repeated speech or accidental touches. For example, if the audiologist quickly says "loud +" twice in succession, the system will execute FIT_BIG_UP the first time, but the second execution will be skipped due to the deduplication mechanism, preventing the loud volume gain from being incorrectly increased twice consecutively.

[0100] This embodiment reduces the false trigger rate and avoids the privacy risks and power consumption issues caused by continuous monitoring by using a non-continuous listening method of pressing and holding to speak. It executes immediately after releasing the hand, with fast response speed and smooth operation, which is suitable for fitting scenarios that require frequent fine-tuning of parameters. The partial / final dual-stage recognition mechanism provides real-time feedback and ensures the accuracy of the final recognition. The deduplication mechanism avoids excessive parameter adjustment caused by repeated execution, improving the security and reliability of the operation.

[0101] The confirmation execution mode allows users to speak freely after clicking the matching button, without needing to hold it down. The instant execution mode requires users to hold down the button to speak, with audio only being recorded while the button is held down. The confirmation execution mode, due to its manual confirmation step, has a relatively slower response time but ensures accuracy. The instant execution mode executes immediately upon release, offering a fast response time and is suitable for high-frequency operations. The confirmation execution mode is suitable for scenarios such as initial fitting, significant parameter adjustments, high ambient noise, and when the audiologist lacks confidence in voice recognition. The instant execution mode is suitable for routine fine-tuning, quiet environments, when the audiologist is already familiar with voice interaction, and when rapid response is required. The system supports dynamic switching between the two modes. audiologists can switch modes at any time during the fitting process as needed. Both modes are accessed through a unified voice floating window module, where the audiologist selects the currently used execution mode. The system loads the corresponding interaction logic and process based on the selected mode, ensuring seamless and consistent mode switching. This dual-mode design balances the accuracy and efficiency of fitting operations, improving the adaptability and user experience of the audiologist's voice digital assistant system.

[0102] In one embodiment, after the dual-mode intent matching module performs text matching between the speech recognition text and the audiologist's closed vocabulary to obtain the audiologist's intent identifier, it is also necessary to verify the audiologist's intent identifier to determine whether the intent identifier is a audiologist-specific intent identifier. This ensures that only professional-level instructions in audiologist mode can enter the subsequent UI same-path execution process, while user-level instructions or invalid instructions are intercepted to avoid accidental operation and unauthorized parameter modification.

[0103] Furthermore, it is verified whether the audiologist's intent identifier is a fitting-specific intent identifier; if the audiologist's intent identifier does not have a fitting-specific prefix, it is determined that the audiologist's intent identifier is not a fitting-specific intent identifier; if the audiologist's intent identifier has a fitting-specific prefix, it is determined that the audiologist's intent identifier is a fitting-specific intent identifier.

[0104] The fitting-specific intent identifiers include at least one of loudness curve adjustment, dynamic compression adjustment, sound quality softening and sharpening, audio processing switch, scene mode switching, and fitting page volume slider adjustment.

[0105] Specifically, the prefix for fitting is FIT_. All intent identifiers starting with FIT_ are fitting-specific intent identifiers, such as FIT_BIG_UP (increase volume curve), FIT_MPO_DOWN (decrease volume curve), FIT_SHARPEN (sharpen), FIT_DENOISE_ON (noise reduction on), FIT_SCENE_OUTDOOR (outdoor mode), FIT_VOL_UP (increase volume on fitting page), etc.

[0106] When the optometrist's intent identifier output by the dual-mode intent matching module starts with FIT_, the system determines that the intent identifier is a fitting-specific intent identifier and allows it to enter the subsequent fitting intent execution module and fitting page UI linkage module execution process.

[0107] When the audiologist's intent identifier does not begin with FIT_, such as user-level intent identifiers like VOL_UP or SCENE_OUTDOOR, or if it is a null value that failed to match, the system determines that the intent identifier is not a fitting-specific intent identifier. In this case, the system does not perform any fitting parameter adjustment operations, but instead performs routing processing based on the intent type: if it is a user-level intent identifier, it is routed to the user command executor for direct device-level BLE writing; if it is a null value that failed to match, it sends a "Not recognized, please try again" message to the audiologist's voice floating window module, prompting the audiologist to re-enter the voice command.

[0108] This embodiment uses the FIT_ prefix identifier as the entry threshold for fitting-specific intents, forming a secondary security check. Even in extreme abnormal situations where the mode filtering of the dual-mode intent matching module occurs, it can ensure that user-level instructions will not mistakenly enter the fitting-level execution path. It clarifies the boundary between fitting-specific intents and user-level intents, which facilitates system maintenance and auditing. It provides deterministic input for subsequent UI execution along the same path, avoiding interface state chaos caused by non-fitting intents entering the fitting page UI linkage module.

[0109] S104. When the optometrist's intent identifier is a fitting-specific intent identifier, the optometrist's intent identifier is routed to the fitting page UI linkage module through the fitting intent execution module.

[0110] In hearing aid fitting scenarios, existing technologies cannot guarantee consistency between the voice execution path and the manual operation path of the fitting UI, leading to potential inconsistencies between the interface display, memory parameters, and Bluetooth transmission. Traditional voice control solutions typically employ a separate voice execution channel, with voice commands directly manipulating underlying data or sending Bluetooth commands, bypassing the business method chain of the fitting page UI. This approach causes the following problems: when the audiologist adjusts parameters via voice, the states of controls such as charts, sliders, and switches on the fitting page UI fail to update synchronously, resulting in inconsistencies between the displayed interface and actual parameters; the fitting data in memory may be out of sync with the parameters written to the hearing aid via Bluetooth, leading to situations where "the interface changes but the device hasn't written" or "the device has written but the interface hasn't refreshed"; and the audiologist cannot trace the history of voice operations through the interface, making parameter changes unauditable.

[0111] To address the aforementioned technical issues, this step utilizes the fitting intent execution module. When the fitting technician's intent identifier is verified as a fitting-specific intent identifier, the intent identifier is routed to the fitting page UI linkage module, ensuring that the execution path after voice triggering is equivalent to manual operation.

[0112] Specifically, the fitting intent execution module, acting as the routing hub between the voice layer and the UI layer, is responsible for receiving the audiologist's intent identifier output by the dual-mode intent matching module and distributing it to the corresponding executor or linkage interface based on the type and attributes of the intent identifier. When the audiologist's intent identifier begins with "FIT_", the fitting intent execution module determines that the intent is a fitting-specific intent and routes it to the performFittingVoiceCommand interface of the fitting page UI linkage module. When the audiologist's intent identifier is a scene-based user intent (such as SCENE_OUTDOOR), the fitting intent execution module routes it to the applySceneSelectionFromVoice method, executing a complete process equivalent to manually clicking the scene drop-down menu. When the audiologist's intent identifier is another executable intent, the fitting intent execution module routes it to the device-level executor for direct BLE parameter writing.

[0113] The routing logic of the fitting intent execution module is implemented through the prefix and category fields of the intent identifier. For fitting-specific intents with the FIT_ prefix, the module first reads the category field of the intent identifier to determine the business dimension to which the intent belongs (such as loudness curve, dynamic compression, sound quality processing, audio switch, etc.), and then packages the intent identifier and category information into a VoiceCommand object, which is then passed to the UI linkage module of the fitting page. The VoiceCommand object contains properties such as intentId (intent identifier), category (category), and params (parameters), providing complete context information for subsequent UI linkage execution.

[0114] The fitting intent execution module is also responsible for permission verification and status checks before execution. Before routing the intent to the fitting page UI linkage module, the module checks whether the current fitting page is in an editable state (e.g., not in parameter loading, not in Bluetooth writing, etc.). If the fitting page is not in an editable state, it sends a message to the fitting therapist's voice floating window module saying "Currently not operable, please try again later," to avoid status conflicts caused by executing voice commands when the system is busy.

[0115] This embodiment uses the routing mechanism of the fitting intent execution module to uniformly guide fitting-specific intents to the fitting page UI linkage module, laying the foundation for subsequent UI execution along the same path. It avoids the problem of inconsistencies between the interface display, memory parameters, and Bluetooth transmission caused by the inconsistency between the voice execution path and the manual operation path. The complete intent context is passed through the VoiceCommand object, which makes it easy for the fitting page UI linkage module to automatically switch panels and call corresponding business methods according to the intent category. The permission verification and status check before execution further improve the stability and security of the system.

[0116] S105. The fitting operation corresponding to the fitter's intent identifier is executed in the target panel through the fitting page UI linkage module to generate target fitting parameters.

[0117] Before performing the fitting operation, it is also necessary to: determine the target panel that matches the audiologist's intent identifier; and switch the current fitting adjustment page to the target panel through the fitting page UI linkage module, wherein the target panel includes a basic adjustment panel, a dynamic compression panel, or an audio processing panel.

[0118] In hearing aid fitting scenarios, the detailed fitting page includes multiple panels such as basic adjustments, dynamic compression, and microphone / audio controls. Frequent switching disrupts the communication flow. On-site, the audiologist needs to repeatedly adjust parameters across different dimensions based on user feedback, such as adjusting the gain curve, then dynamic compression, and finally activating noise cancellation. Traditionally, the audiologist needs to manually click panel switching buttons, switching back and forth between different panels, which is cumbersome and interrupts communication with the wearer. If voice commands do not automatically switch to the corresponding panel, the audiologist still needs to switch manually, failing to fully utilize the efficiency advantages of voice control.

[0119] To address the aforementioned technical issues, before the fitting operation is performed by the UI linkage module on the fitting page, the system first identifies a target panel that matches the audiologist's intent. Then, it automatically switches the current fitting adjustment page to that target panel, allowing the audiologist to perform parameter adjustments in the correct panel context without manual intervention.

[0120] Specifically, the UI linkage module on the fitting page maintains a mapping table between intent identifiers and target panels. This mapping table defines the target panels corresponding to various fitting-specific intent identifiers. The mapping relationship is as follows: the target panels corresponding to loudness curve adjustment intent identifiers (FIT_BIG_UP / DOWN, FIT_MID_UP / DOWN, FIT_SMALL_UP / DOWN, FIT_LANG_UP / DOWN, FIT_LOUDNESS_UP / DOWN) and sound quality softening and sharpening intent identifiers (FIT_SOFTEN, FIT_SHARPEN) are basic adjustment panels; the target panels corresponding to dynamic compression adjustment intent identifiers (FIT_MPO_UP / DOWN, FIT_GEAR_UP / DOWN) are dynamic compression adjustment panels. Shrink the panel; the target panel corresponding to the audio processing switch intent markers (FIT_BEAM_ON / OFF, FIT_ORIGINAL_ON / OFF, FIT_DENOISE_ON / OFF, FIT_WIND_ON / OFF, FIT_TRANSIENT_ON / OFF, FIT_SQUEAL_ON / OFF) is the audio processing panel; the scene mode switching intent markers (FIT_SCENE_*) and the fitting page volume slider adjustment intent markers (FIT_VOL_UP / DOWN) do not trigger panel switching, and are executed directly in the current panel.

[0121] Once the target panel is determined, the fitting page UI linkage module switches the current fitting adjustment page to the target panel by calling the panel switching method. Specifically, when the target panel is the basic adjustment panel, the `voiceEnsureBasicMode()` method is called. This method is exactly the same as the method called when manually clicking the "Basic Adjustment" tab, synchronously updating UI elements such as the selected state of the top navigation bar, the display of the panel content area, and the state of the bottom toolbar. When the target panel is the dynamic compression panel, the `voiceEnsureDynamicMode()` method is called. This method is exactly the same as the method called when manually clicking the "Dynamic Compression" tab, synchronously loading the dynamic compression parameter table, refreshing the MPO and gear display. When the target panel is the audio processing panel, the `voiceEnsureAudioMode()` method is called. This method is exactly the same as the method called when manually clicking the "Audio Processing" tab, synchronously loading the audio processing switch state and parameter settings.

[0122] During panel switching, the fitting page UI linkage module inherits the currently selected ear type and scene type, ensuring that the parameters displayed on the panel after the switch are consistent with those before the switch. For example, when the left ear and outdoor scene are currently selected, after switching from the basic adjustment panel to the dynamic compression panel, the dynamic compression panel will still display the MPO and gear parameters for the left ear outdoor scene, without any loss of ear type or scene context.

[0123] This embodiment automatically maps intent identifiers to target panels, allowing audiologists to smoothly execute voice commands across different panels without manually clicking panel switching buttons, maintaining the continuity of fitting communication. The panel switching method is completely consistent with manual operation, ensuring the correctness and consistency of the UI state. Inheriting the current ear type and scene context avoids parameter display confusion after switching, improving operational efficiency.

[0124] In hearing aid fitting scenarios, existing technologies cannot guarantee consistency between the voice execution path and the manual operation path via the fitting UI, potentially leading to inconsistencies between the interface display, memory parameters, and Bluetooth commands. Traditional voice control solutions bypass the UI layer to directly manipulate data or send Bluetooth commands, causing a disconnect between the interface state and actual parameters. The audiologist cannot verify the results of voice operations through the interface, and parameter changes are untraceable. Furthermore, parameter adjustments under binaural and multi-scenario conditions require precise positioning of parameter arrays specific to the ear and scenario. If voice operations cannot inherit the current context, it may result in incorrect parameter modification locations, affecting fitting accuracy.

[0125] To address the aforementioned technical issues, this step utilizes the UI linkage module on the fitting page to invoke the same business method chain as the manual UI operation in the target panel. Based on the currently selected ear identifier and scenario identifier, it executes the fitting operation corresponding to the audiologist's intent identifier, updates the fitting parameter array corresponding to the ear identifier and scenario, and generates the target fitting parameters.

[0126] Furthermore, based on the currently selected ear identifier and scene identifier, the same business method chain as the manual UI operation is called through the fitting page UI linkage module to perform the fitting operation corresponding to the audiologist's intent identifier in the target panel, update the fitting parameter array corresponding to the ear identifier and the scene identifier, and generate the target fitting parameters.

[0127] First, the current context is read. The fitting page UI linkage module reads the values ​​of currentEarType (current ear type identifier, such as left ear, right ear, or both ears) and sceneType (current scene identifier, such as general, outdoor, noisy, quiet, TV, conversation) to determine the target ear type and target scene for the current operation.

[0128] Secondly, the parameter array is located. Based on currentEarType and sceneType, the corresponding fitting parameter array is located from the fitting data structure in memory. The fitting data structure is organized as a three-dimensional array. The first dimension is ear type (left ear / right ear), the second dimension is scene (general / outdoor / noisy / quiet / TV / conversation), and the third dimension is parameter type (small / mid / big gain curve, MPO, gear, on / off state, etc.). For example, when currentEarType is left ear and sceneType is outdoor, the system locates the parameter array corresponding to the outdoor scene for the left ear.

[0129] Then, the fitting operation is executed by calling the business method chain. Based on the medic's intent, the UI-linked module on the fitting page calls the same business method chain as the manual UI operation, executing the corresponding fitting operation in the target panel. Each method in the business method chain is completely identical to the method called when manually clicking a UI control, ensuring the equivalence of the execution path.

[0130] Taking the "sharpen" command as an example, when the optician says "sharpen" on the fitting page, the recognition software, after text normalization, hits FIT_SHARPEN. The fitting page UI linkage module first calls voiceEnsureBasicMode() to switch to the basic adjustment panel (if it is not currently in the basic adjustment panel), such as... Figure 3 As shown, the basic adjustment interface includes a gain curve graph and a basic adjustment panel. The gain curve graph is located in the upper half of the interface, with the horizontal axis representing frequency (Hz) and the vertical axis representing gain (dB), displaying four gain curves: Mpo, soft, medium, and loud. The basic adjustment panel is located in the lower half of the interface, containing a volume adjustment slider, a loudness adjustment area (including buttons for loud, medium, soft, and speech), and a sound quality area (including soften and sharpen buttons). Then, `voiceApplySoundQuality(sharpen:true)` is called. This method is similar to manually clicking... Figure 3 The method called when using the "Sharpen" button as shown in the right figure is exactly the same: select the corresponding small / mid / big array according to currentEarType and sceneType, call HearingAdjustTools.getSharpenData to calculate the new curve, update the gain curve parameter array corresponding to the ear type and scene, and refresh the chart in real time to display the new gain curve.

[0131] Taking the "MPO+" command as an example, the optician says that "MPO+" hits FIT_MPO_UP. The fitting page UI linkage module first calls voiceEnsureDynamicMode() to switch to the dynamic compression panel (if it is not currently in the dynamic compression panel), such as... Figure 4As shown, the dynamic compression panel includes a gain curve chart and a dynamic compression parameter adjustment interface. The gain curve chart is located in the upper half of the interface, displaying the gain curve for the current earphone and scene. The dynamic compression parameter adjustment interface is located in the lower half of the interface, displaying the MPO value and adjustment buttons (-5, -1, +1, +5) for each frequency point (e.g., 125Hz, 187Hz, 357Hz) in a table format. Then, voiceApplyMPO(step:1) is called. This method is similar to manually clicking... Figure 4 The method called when adding an MPO is exactly the same: it reads the current MPO value based on currentEarType and sceneType, increases it by a preset step size (e.g., 1dB), updates the corresponding ear type and scene MPO parameters, and refreshes the dynamic compression table, which displays the new MPO value in real time. Figure 6 As shown, in the dynamic compression panel, audiologists can click the MPO drop-down menu to switch between different compression parameters such as MPO, soft, medium, and loud to view and adjust the compression configuration under different parameter dimensions.

[0132] Taking the "Turn on noise cancellation" command as an example, the audiologist says "Turn on noise cancellation" triggers FIT_DENOISE_ON. The UI module on the fitting page first calls voiceEnsureAudioMode() to switch to the audio processing panel, such as... Figure 5 As shown, the microphone audio processing panel includes multiple audio processing switches and sliders, including a beamformer switch, a feedback suppression switch, a WDRC switch, a noise reduction switch and slider (currently at level 10), a wind noise switch and slider (currently at level 3), a transient noise switch and slider (currently at level 1), and a sound processing switch and slider (currently at level 1). Then, `voiceApplyDenoise(enable:true)` is called. This method is similar to manually clicking... Figure 5 The methods called when switching noise reduction on and off are exactly the same: setting the noise reduction switch state to "on" for the corresponding ear and scene, updating the audio processing switch array, refreshing the switch display, and updating the switch state to "on" in real time.

[0133] Taking the "Switch to Outdoor Mode" command as an example, the audiologist says "Switch to Outdoor Mode" hits FIT_SCENE_OUTDOOR. The UI linkage module on the fitting page calls applySceneSelection(type:2), which is exactly the same method called when manually clicking the scene drop-down menu to select Outdoor Mode. It synchronously updates the scene button title to "Outdoor," reads the gain curve data corresponding to the outdoor scene from memory to refresh the top chart, reads the dynamic compression parameters to refresh the dynamic compression table, and reads the audio header parameters to refresh the audio processing panel. Please refer to... Figure 7 and Figure 8 , Figure 7 This application provides a schematic diagram of switching between different panels. Figure 8 This is a schematic diagram illustrating the switching between different modes provided in this application. For example... Figure 7 As shown, audiologists can switch between three panels—Basic Adjustment, Wide Dynamic Range Compression, and Microphone / Audio—by clicking the drop-down menu on the panel to access different dimensions of fitting parameters. Figure 8 As shown, audiologists can switch between six scene modes—General Mode, Outdoor Mode, Noisy Mode, Quiet Mode, TV Mode, and Conversation Mode—by clicking the mode switching drop-down menu to adapt to different acoustic environments.

[0134] Finally, the target fitting parameters are generated. After the business method chain is executed, the updated fitting parameter array becomes the target fitting parameters. This parameter array is stored in memory and is completely consistent with the memory state after manual operation, ensuring the consistency of the memory data.

[0135] In this embodiment, by calling the same business method chain as the manual UI operation, the equivalence between the voice execution path and the manual operation path is ensured, achieving traceability and consistency of "voice-based click". Based on the current ear identification and scene context, the parameter array is located, and only the parameters of the corresponding ear identification and scene are modified, avoiding incorrect parameter modification locations. This is particularly suitable for binaural fitting and multi-scene comparison debugging. Each method in the business method chain reuses the manual operation method, eliminating the need to develop additional voice-specific logic, thus reducing development costs and maintenance difficulty. After the target fitting parameters are generated, they are stored in memory, consistent with the memory state after the manual operation, providing a reliable data foundation for subsequent Bluetooth parameter distribution.

[0136] S106. The target fitting parameters are sent to the hearing aid through the Bluetooth parameter sending module to update the current fitting parameters of the hearing aid to the target fitting parameters.

[0137] In hearing aid fitting scenarios, ensuring synchronization between the interface, memory-based fitting data, and Bluetooth transmission after voice triggering is a technical challenge that current technologies struggle to address. Traditional voice control solutions bypass the UI layer and directly send Bluetooth commands, leading to inconsistencies between memory data and Bluetooth-written data, or between the parameters displayed on the interface and those actually written. Furthermore, scene switching operations involve the synchronous updating of multiple parameters; incomplete or incorrect Bluetooth transmission can cause hearing aid malfunctions.

[0138] To address the aforementioned technical issues, this step uses a Bluetooth parameter transmission module to write the target fitting parameters generated by the fitting page UI linkage module into the hearing aid via a Bluetooth link, ensuring consistency between the interface display, the memory fitting data, and the Bluetooth transmission.

[0139] Specifically, the Bluetooth parameter delivery module receives the write trigger signal from the fitting page UI linkage module, reads the updated target fitting parameters from memory, and writes the parameters to the hearing aid via the Bluetooth communication link. The write trigger signal is automatically issued by the fitting page UI linkage module after the business method chain is completed, and is the same as the signal triggered by clicking the "write" button after manual operation.

[0140] Taking the "sharpen" command as an example, after the UI linkage module on the fitting page refreshes the top chart and updates the gain curve parameter array in memory via `changeSceneData`, it asynchronously calls the `clickWrite()` method. The `clickWrite()` method is exactly the same as the method called when manually clicking the "Write" button. It packages the updated small / mid / big gain curve parameter array for the left ear outdoor scene (assuming the current context) in memory into a Bluetooth data packet and sends it to the hearing aid via the BLE communication link. After receiving the data packet, the hearing aid parses and updates its internal parameters, returning a write confirmation signal. Upon receiving the confirmation signal, the Bluetooth parameter delivery module sends a message "Executed: Sharpening" to the audiologist's voice floating window module, completing the entire voice command execution process.

[0141] Taking the "Outdoor Mode" scene switching as an example, after the UI linkage module on the fitting page synchronously updates the scene button title, gain curve, dynamic compression table, and audio header parameters via applySceneSelection, it writes the scene configuration via the Bluetooth API and calls clickWrite(). The Bluetooth parameter delivery module packages the complete parameter configuration corresponding to the outdoor scene (including gain curve, dynamic compression parameters, audio processing switch status, etc.) into a scene configuration data packet and sends it to the hearing aid via the Bluetooth link. After receiving the packet, the hearing aid switches to the outdoor scene mode, loads the corresponding parameter configuration, and returns a scene switching confirmation signal. After receiving the confirmation signal, the Bluetooth parameter delivery module sends feedback to the audiologist's voice floating window module: "Executed: Outdoor Mode".

[0142] For scene switching intents, the Bluetooth parameter delivery module calls the same scene switching method as the scene dropdown menu selection, synchronously updating the gain curve, dynamic compression table, audio header parameters, and Bluetooth configuration. Specifically, after updating memory parameters and refreshing the UI, the applySceneSelection method calls the Bluetooth API's writeSceneConfig method to write the complete parameter configuration corresponding to the new scene to the hearing aid. The writeSceneConfig method is completely consistent with the Bluetooth write method called when manually switching scenes, ensuring the integrity and consistency of scene switching.

[0143] The Bluetooth parameter delivery module also supports a retry mechanism for write failures. If the initial Bluetooth write fails to receive a confirmation signal from the hearing aid or receives an error response, the module automatically retryes up to three times, with each retry spaced 500ms apart. If the retry still fails, it sends a message to the audiologist's voice pop-up module stating "Write failed, please check the Bluetooth connection," and retains the target fitting parameters in memory. The write process can be triggered again once the Bluetooth connection is restored. This retry mechanism ensures the reliability of parameter delivery and avoids parameter loss due to momentary Bluetooth signal interference.

[0144] This embodiment writes the updated fitting parameters to the hearing aid via Bluetooth, ensuring consistency between the interface display, the in-memory fitting data, and the Bluetooth transmission. This solves the technical problem of asynchronous communication between the interface, in-memory fitting data, and Bluetooth transmission after voice triggering. By reusing the clickWrite() method after manual operation, the equivalence of the Bluetooth transmission path and the manual operation path is ensured. The asynchronous writing mechanism avoids blocking UI refresh via Bluetooth communication, improving system response speed. The complete parameter configuration transmission for scene switching intents ensures the integrity and correctness of hearing aid scene mode switching. The retry mechanism for writing failures improves the reliability of parameter transmission and reduces the risk of parameter loss due to communication problems.

[0145] This embodiment provides a hearing aid fitting parameter configuration method based on speech recognition. This method uses a speech acquisition and recognition module to input commands via voice, replacing traditional manual touch operation. This allows audiologists to adjust parameters even when their hands are occupied observing user reactions and recording feedback, avoiding operational interruptions caused by multitasking. A dual-mode intent matching module calls the audiologist's closed vocabulary, isolating user-level and audiologist-level commands. This ensures that the audiologist's dedicated intent identifier can only match fine-tuning commands, preventing misoperation by non-professionals. Furthermore, the fitting intent execution module routes the audiologist's intent identifier to the fitting page UI linkage module, directly executing the corresponding fitting operation and generating target fitting parameters in the target panel. This eliminates the manual steps of frequently switching between multiple function panels and searching for corresponding controls. Finally, the target fitting parameters are sent to the hearing aid via a Bluetooth parameter delivery module, completing the parameter update. This allows audiologists to complete fine-tuning parameters without interrupting the current workflow, improving fitting efficiency.

[0146] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for configuring hearing aid fitting parameters based on speech recognition, characterized in that, An audio-visual digital assistant system for opticians is applied, comprising a voice acquisition and recognition module, a dual-mode intent matching module, a optician-specific vocabulary module, an optician intent execution module, an optician page UI linkage module, and a Bluetooth parameter transmission module; the method includes: The voice input signal of the optician is acquired by the voice acquisition and recognition module, and the voice input signal is recognized to obtain the voice recognition text. Call the closed terminology library for audiologists that matches the audiologist mode in the dedicated terminology library module for audiologists; Based on the aforementioned optometrist mode, the dual-mode intent matching module performs text matching between the speech recognition text and the optometrist closed vocabulary to obtain the optometrist intent identifier. When the optometrist's intent identifier is a fitting-specific intent identifier, the fitting intent execution module routes the optometrist's intent identifier to the fitting page UI linkage module. The fitting operation corresponding to the audiologist's intent identifier is executed in the target panel through the fitting page UI linkage module, and the target fitting parameters are generated. The target fitting parameters are sent to the hearing aid via the Bluetooth parameter sending module to update the current fitting parameters of the hearing aid to the target fitting parameters.

2. The hearing aid fitting parameter configuration method based on speech recognition according to claim 1, characterized in that, Before the step of executing the fitting operation corresponding to the optometrist's intent identifier in the target panel through the fitting page UI linkage module and generating the target fitting parameters, the method further includes: Identify the target panel that matches the optometrist's intent identifier; The fitting page UI linkage module switches the current fitting adjustment page to the target panel, wherein the target panel includes a basic adjustment panel, a dynamic compression panel, or an audio processing panel.

3. The hearing aid fitting parameter configuration method based on speech recognition according to claim 1, characterized in that, Before routing the optician's intent identifier to the fitting page UI linkage module through the fitting intent execution module when the optician's intent identifier is a fitting-specific intent identifier, the method further includes: Verify whether the optometrist's intent identifier is a fitting-specific intent identifier; When the optometrist's intent identifier is not identified by a fitting-specific prefix, it is determined that the optometrist's intent identifier is not a fitting-specific intent identifier. When the optometrist's intent identifier is identified by a fitting-specific prefix, the optometrist's intent identifier is determined to be a fitting-specific intent identifier.

4. The method for configuring hearing aid fitting parameters based on speech recognition according to claim 3, characterized in that, The fitting-specific intent identifiers include at least one of the following: loudness curve adjustment, dynamic compression adjustment, sound quality softening and sharpening, audio processing switch, scene mode switching, and fitting page volume slider adjustment.

5. The method for configuring hearing aid fitting parameters based on speech recognition according to claim 1, characterized in that, The step of executing the fitting operation corresponding to the optician's intent identifier in the target panel through the fitting page UI linkage module, and generating target fitting parameters, includes: Based on the currently selected ear identifier and scene identifier, the same business method chain as the manual UI operation is called through the fitting page UI linkage module. The fitting operation corresponding to the audiologist's intent identifier is executed in the target panel, the fitting parameter array corresponding to the ear identifier and scene identifier is updated, and the target fitting parameters are generated.

6. The hearing aid fitting parameter configuration method based on speech recognition according to claim 1, characterized in that, The dual-mode intent matching module filters the unified instruction catalog through the instruction category field and the audience field, thereby achieving the isolation of the word library between the user mode and the optician mode.

7. The hearing aid fitting parameter configuration method based on speech recognition according to claim 1, characterized in that, The optometrist voice digital assistant system also includes an optometrist voice floating window module; the method further includes: The audiologist's voice floating window module, attached to the fitting adjustment page, receives the audiologist's voice switch trigger operation and press-to-talk operation. After the voice input signal is collected, the audiologist's voice floating window module provides feedback on the execution result to the audiologist.

8. The method for configuring hearing aid fitting parameters based on speech recognition according to claim 7, characterized in that, Before routing the optician's intent identifier to the fitting page UI linkage module through the fitting intent execution module when the optician's intent identifier is a fitting-specific intent identifier, the method further includes: When the optician's voice digital assistant system is configured in confirmation execution mode, the voice acquisition and recognition module acquires the optician's continuous voice input signal and performs voice recognition on the voice input signal to obtain voice recognition text containing at least one voice command. The voice-recognized text is displayed through the optician's voice floating window module, and the optician's confirmation operation on the voice-recognized text is received. Upon receiving the confirmation operation, the dual-mode intent matching module performs text matching between the speech recognition text and the optician's closed vocabulary to obtain the optician's intent identifier.

9. The method for configuring hearing aid fitting parameters based on speech recognition according to claim 7, characterized in that, Before routing the optician's intent identifier to the fitting page UI linkage module through the fitting intent execution module when the optician's intent identifier is a fitting-specific intent identifier, the method further includes: When the optician's voice digital assistant system is configured to instant execution mode, the optician's press-to-talk operation is detected through the optician's voice floating window module. During the press-to-talk operation, the voice input signal of the optician is acquired in real time through the voice acquisition and recognition module, and the voice input signal is recognized to obtain real-time recognized text. After the press-to-talk operation is detected to be released, the dual-mode intent matching module performs text matching between the real-time recognized text and the audiologist's closed vocabulary to obtain the audiologist's intent identifier.

10. The method for configuring hearing aid fitting parameters based on speech recognition according to claim 1, characterized in that, Based on the optometrist mode, the dual-mode intent matching module performs text matching between the speech recognition text and the optometrist closed vocabulary to obtain the optometrist intent identifier, including: The speech-recognized text is then normalized to obtain normalized text; Based on the aforementioned optometrist model, the optometrist closed-loop terminology is filtered to obtain a set of usable rules; Based on a preset matching strategy, each rule in the available rule set is matched with the normalized text, and the matching confidence of each rule is calculated. When the highest confidence level is greater than or equal to the preset confidence threshold, the corresponding optometrist intent identifier is output.

Images