Method, System, and Computer-Readable Storage Medium for Setting Initial Conditions to Limit Search Candidates in Speech Processing

Abstract

The present invention relates to the field of voice signal processing and voice-based information processing. A method for setting initial conditions for voice processing according to an embodiment of the present invention may include: receiving a processing target input comprising at least one of a voice signal or a unit sequence derived from voice; determining at least one of a target processing interval, a boundary, and a processing target unit for the input, and determining whether to apply a standard procedure loop and whether to apply an exception path; calculating initial condition information based on the input and the determination result; setting a VC anchor, which is an initial state reference point, to an initial state based on the initial condition information according to the determination result regarding whether to apply a standard procedure loop and whether to apply an exception path, or setting the VC anchor setting to be conditionally bypassed; generating, setting, or updating constraint or candidate restriction information for limiting search candidates for subsequent processing, and providing the constraint or candidate restriction information to subsequent processing or to reference and apply it during the search process of subsequent processing; and providing a preprocessing result including the initial condition information and the constraint or candidate restriction information to subsequent processing.

Description

Technology Field

[0001] The present invention relates to the field of voice signal processing and voice-based information processing, and more specifically, to a method, system, and storage medium for explicitly setting initial conditions for an input at a stage prior to inputting it to subsequent processing, and providing constraint or candidate restriction information based on said initial conditions.

[0002] The present invention is based on various processing target inputs that can be provided for subsequent processing, such as a voice signal received as input and / or information regarding unit sequences, processing intervals, and boundaries derived from the voice, and enables the configuration of initial conditions that can be referenced in subsequent processing through an initialization procedure for said inputs. The processing target inputs may be included together in a combined form of a voice signal and a unit sequence provided to correspond to said voice signal.

[0003] The above initialization procedure may generate initial condition information and search constraint or candidate restriction information to provide for subsequent processing, and depending on the implementation, may also provide record or display information regarding said information. As used herein, 'record or display information' may be implemented in various forms such as output, storage, display, log, status column, etc., but means provided as a unit of information that can be referenced in subsequent processing (not limited thereto). The present invention may be applied to various voice processing procedures and may be implemented in various system / device / program forms without being limited by a specific implementation method, specific processing environment, or specific output format. The present invention may be implemented in various terminals such as smartphones, wearables, radio terminals, and vehicle / equipment-mounted terminals, and may be applied to contribute to the stable operation of voice-based control, guidance, alarm, and reporting functions even in environments with high background noise or significant fluctuations in communication quality, such as work sites (not limited thereto). Background Technology

[0004] In the fields of speech signal processing and speech-based information processing, preprocessing and initialization of input speech can affect the operation and results of subsequent processing. Generally, input speech may undergo preprocessing and initialization before subsequent processing, and the final result can be produced by performing search and decision-making based on the provided information during subsequent processing. In this case, the operation of subsequent processing can be influenced by the quality of boundary information, initial state-related information, and constraint information provided in the preceding stage.

[0005] Typically, the above preprocessing and initialization can be implemented or assisted through idle interval detection, alignment-based processing, or exploration of the neural network decoding process. Recently, voice interfaces in various devices such as vehicles, TVs, and smart devices are expanding beyond simple question-and-answer to agent-based interactions that include the execution of transactions such as reservations, payments, and ticketing.

[0006] In such environments, a single misrecognition can lead to financial loss, reduced reliability, or safety issues, which reduces the allowable margin of error for subsequent processing and increases the importance of setting initial conditions and limiting candidates in the upstream stage. In particular, in industrial / work environments (e.g., manufacturing, logistics, vehicle / mobility, home / mobile terminal environments, construction, etc.), high noise, reverberation, irregular interference, and multi-speaker environments are frequent, and variations in voice input quality can be significant due to radio communication, broadcasting, and the wearing of protective equipment. Under these conditions, the likelihood of misrecognizing safety-related utterances such as short commands, warnings, and acknowledgments may increase; therefore, there may be an increased need to structure initial conditions for reference by subsequent voice recognition or processing steps on an input-by-input basis and to control the search space (but is not limited thereto). In some implementations, preprocessing and initialization procedures such as segmentation, boundary detection, normalization, initial alignment, or initial hypothesis setting are performed in a rule-based or model-based manner, and the results can be indirectly reflected in subsequent processing.

[0007] Inputs with unclear boundaries or complex input conditions may exist, which can increase the number of candidates considered in subsequent processing. Accordingly, the importance of setting initial pre-processing conditions and limiting candidates may increase to ensure real-time performance, resource efficiency, and stable operation.

[0008] Furthermore, if initial conditions are not explicitly defined or provided, or are only implicitly reflected, the starting state of subsequent processing becomes sensitive to input fluctuations, leading to unstable results. Additionally, small errors in pre-processing judgments can propagate to subsequent stages, potentially resulting in performance degradation. Moreover, changes in the operating environment or target data characteristics often require the resetting, additional adjustment, or correction of rules, feature selections, and thresholds, which can increase maintenance costs and limit the scope of application.

[0009] Accordingly, there is a need for technology that can more explicitly set initial conditions in the stage prior to subsequent processing and systematically provide constraint and candidate restriction information available to the subsequent stage. Prior art literature

[0010] Korean Published Patent Application No. 10-2014-0092960 (July 25, 2014) The problem to be solved

[0011] The present invention aims to mitigate the problem in which the search burden increases and errors in pre-processing judgments can propagate to subsequent stages as initial conditions are implicitly or unstably reflected in the stage prior to subsequent processing.

[0012] To this end, the core of the present invention lies in providing an initialization procedure that explicitly sets the initial state and initial conditions for a target processing section at a stage prior to being input to subsequent processing, and enables the result to be utilized in subsequent processing. The input may include processing target inputs that can be provided to subsequent processing, such as voice signals and / or unit sequences derived from voice, processing sections, and boundaries. means of solving the problem

[0013] A method for setting initial conditions for voice processing according to an embodiment of the present invention for solving the above problem may include: receiving a processing target input including at least one of a voice signal or a unit sequence derived from voice; determining at least one of a target processing interval, a boundary, and a processing target unit for the input, and determining whether to apply a standard procedure loop and whether to apply an exception path; calculating initial condition information based on the input and the determination result; setting a VC anchor, which is an initial state reference point, to an initial state based on the initial condition information according to the determination result regarding whether to apply a standard procedure loop and whether to apply an exception path, or setting the VC anchor setting to be conditionally bypassed; generating, setting, or updating constraint or candidate restriction information to limit search candidates for subsequent processing, and providing the constraint or candidate restriction information to subsequent processing or to reference and apply it during the search process of subsequent processing; and providing a preprocessing result including the initial condition information and the constraint or candidate restriction information to subsequent processing.

[0014] In a method according to an embodiment of the present invention for solving the above problem, when the application of the standard procedure loop is determined, at least some of the steps may be configured to be implemented in a sequence including setting a VC anchor, determining a consonant type, determining a vowel type, and terminating the process.

[0015] In a method according to an embodiment of the present invention for solving the above problem, when the application of the standard procedure loop is determined, the result of combining at least some of the steps can be configured to be implemented as a finite set of loops.

[0016] In a method according to an embodiment of the present invention for solving the above problem, when the application of the exception path is determined, the VC anchor setting may be conditionally bypassed, and record or display information reflecting the exception state is generated, and then the process may be transitioned to a subsequent process.

[0017] In a method according to an embodiment of the present invention for solving the above problem, the termination process is configured to result in either a vowel termination or a termination by a termination marker, and the termination by the termination marker may be configured not to presuppose the existence of a specific consonant component or to be limited to a specific consonant.

[0018] In a method according to an embodiment of the present invention for solving the above problem, the constraint or candidate restriction information is configured to be provided in a form that may include, for example, a list of allowed or prohibited candidates, a state transition restriction mask, a set of candidate indices, metadata regarding candidate priority or weight, score bias or penalty, or control information regarding a threshold, and the constraint or candidate restriction information may be configured to restrict the activation of a search candidate or state transition path in subsequent processing.

[0019] In a method according to an embodiment of the present invention for solving the above problem, at least a portion of the step of generating, setting, or updating and providing the constraint or candidate restriction information and the determination step may be configured to be performed by hardware accelerator logic.

[0020] In a method according to an embodiment of the present invention for solving the above problem, the preprocessing result may be configured to optionally include record or display information indicating a surface realization of a consonant stage.

[0021] In a method according to an embodiment of the present invention for solving the above problem, the preprocessing result may be configured to optionally further include record or display information indicating the progression position from a preceding vowel component to a succeeding vowel component when the vowel is determined to be VV.

[0022] In a method according to an embodiment of the present invention for solving the above problem, the preprocessing result may be configured to include a representative result in which a processing unit corresponding to a syllable-ending component is mapped to a representative value.

[0023] In a method according to an embodiment of the present invention for solving the above problem, language-specific differences may be applied by separating them into language-specific parameter sets including language-specific unit inventories and transition and combination constraints, and may be configured to be expandable without changing the steps constituting the method by adding or modifying the language-specific parameter sets.

[0024] In a method according to an embodiment of the present invention for solving the above problem, at least one of the determination step and the step of calculating the initial condition information may be configured to be performed by rule-based processing, statistical model-based processing, or a combination thereof.

[0025] In a method according to an embodiment of the present invention for solving the above problem, at least one of the determination step and the step of calculating the initial condition information may be configured to select and perform at least one of rule-based processing and model-based processing depending on the input characteristics or execution environment.

[0026] In a method according to an embodiment of the present invention for solving the above problem, when the application of the standard procedure loop is determined, the initial condition information includes preliminary articulation setting information that affects the articulation intensity of a subsequent consonant, and the step of setting the VC anchor to an initial state may be configured to reflect the preliminary articulation setting information.

[0027] In a method according to an embodiment of the present invention for solving the above problem, at least some of the determination step, the step of calculating the initial condition information, and the step of generating, setting, or updating and providing the constraint or candidate restriction information may be distributed so that they are performed in an on-device environment and the remaining parts are performed in a server environment.

[0028] In a method according to an embodiment of the present invention for solving the above problem, the step of calculating the initial condition information may be configured to be performed by referring to at least a portion of the preprocessing result provided for the immediate preceding processing unit when calculating the initial condition information for the next processing unit.

[0029] An initial condition setting system for voice processing according to an embodiment of the present invention for solving the above problem may include: an input receiving unit configured to receive a processing target input including at least one of a voice signal or a unit sequence derived from voice; a determination unit configured to determine at least one of a target processing interval, a boundary, and a processing target unit for the input, and to determine whether a standard procedure loop is applied and whether an exception path is applied; an initial condition information generating unit configured to calculate initial condition information based on the input and the determination result; an initial state setting unit configured to set a VC anchor, which is an initial state reference point, to an initial state based on the initial condition information, or to set the VC anchor setting to bypass the conditional state, according to the determination result regarding whether a standard procedure loop is applied and whether an exception path is applied; a constraint or candidate restriction information generating and providing unit configured to generate, set, or update constraint or candidate restriction information for limiting search candidates for subsequent processing, and to provide the constraint or candidate restriction information to subsequent processing or to reference and apply it during the search process of subsequent processing; and an output unit or providing unit configured to provide a preprocessing result including the initial condition information and the constraint or candidate restriction information to subsequent processing.

[0030] In a system according to an embodiment of the present invention for solving the above problem, at least a portion of the determination unit and the constraint or candidate restriction information generation and provision unit may be configured to be performed by hardware accelerator logic.

[0031] In a system according to an embodiment of the present invention for solving the above problem, when the application of the standard procedure loop is determined, at least one of the initial state setting unit and the determination unit may be configured to perform step-by-step processing including VC anchor setting, consonant type determination, vowel type determination, and termination processing.

[0032] In a system according to an embodiment of the present invention for solving the above problem, when the application of the standard procedure loop is determined, the result of the combination of step-by-step processing including VC anchor setting, consonant type determination, vowel type determination, and termination processing can be configured to be implemented as a finite set of loops.

[0033] In a system according to an embodiment of the present invention for solving the above problem, when the application of the exception path is determined, the initial state setting unit may be configured to conditionally bypass the VC anchor setting, generate record or display information reflecting the exception state, and then transition to subsequent processing.

[0034] In a system according to an embodiment of the present invention for solving the above problem, at least one of the determination unit, the initial state setting unit, and the constraint or candidate restriction information generation and provision unit is configured such that the termination process results in either a vowel termination or a termination by a termination marker, and the termination by the termination marker may be configured such that it presupposes the existence of a specific consonant component or is not limited to a specific consonant.

[0035] In a system according to an embodiment of the present invention for solving the above problem, the constraint or candidate restriction information generation and provision unit is configured to generate and provide the constraint or candidate restriction information in a form that may include, for example, a list of allowed or prohibited candidates, a state transition restriction mask, a set of candidate indices, metadata regarding candidate priority or weight, score bias or penalty, or control information regarding threshold values, and the constraint or candidate restriction information may be configured to restrict the activation of a search candidate or state transition path in subsequent processing.

[0036] In a system according to an embodiment of the present invention for solving the above problem, the output unit or the providing unit may be configured to include record or display information indicating a surface realization of a consonant stage in the preprocessing result.

[0037] In a system according to an embodiment of the present invention for solving the above problem, the output unit or the providing unit may be configured to selectively include record or display information indicating the progression position from a preceding vowel component to a succeeding vowel component in the preprocessing result when the vowel is determined to be VV.

[0038] In a system according to an embodiment of the present invention for solving the above problem, at least one of the constraint or candidate restriction information generation and provision unit and the output unit or provision unit may be configured to include a representative result in which a processing unit corresponding to a syllable-ending component is mapped to a representative value in the preprocessing result.

[0039] In a system according to an embodiment of the present invention for solving the above problem, at least one of the determination unit and the constraint or candidate restriction information generation and provision unit may be configured to apply a language-specific parameter set including a language-specific unit inventory and transition and combination constraint conditions so that at least one of the determination and the generation of the constraint or candidate restriction information may vary by language.

[0040] In a system according to an embodiment of the present invention for solving the above problem, at least one of the determination unit and the initial condition information generation unit may be configured to be performed by rule-based processing, statistical model-based processing, or a combination thereof.

[0041] In a system according to an embodiment of the present invention for solving the above problem, at least one of the determination unit and the initial condition information generation unit may be configured to select and perform at least one of rule-based processing and model-based processing depending on input characteristics or execution environment.

[0042] In a system according to an embodiment of the present invention for solving the above problem, when the application of the standard procedure loop is determined, the initial condition information generating unit may be configured to generate initial condition information including preliminary articulation setting information that affects the articulation intensity of a subsequent consonant, and the initial state setting unit may be configured to set the VC anchor to an initial state by reflecting the preliminary articulation setting information.

[0043] In a system according to an embodiment of the present invention for solving the above problem, at least some of the determination unit, the initial condition information generation unit, and the constraint or candidate restriction information generation and provision unit may be configured to be placed in an on-device environment and the remaining parts placed in a server environment to communicate with each other and perform distributed execution, and the mutual communication may include the transmission of at least one of the preprocessing result or a metadata record or a log record regarding the preprocessing result.

[0044] In a system according to an embodiment of the present invention for solving the above problem, the initial condition information generating unit may be configured to refer to at least a portion of the preprocessing result provided for the immediate previous processing unit when calculating initial condition information for the next processing unit.

[0045] A non-transient computer-readable storage medium according to an embodiment of the present invention for solving the above problem may store a program instruction that causes the method to be performed when executed by a processor.

[0046] In a storage medium according to an embodiment of the present invention for solving the above problem, at least one of an intermediate representation including a feature column, a state column or metadata and the preprocessing result is configured to be loaded or updated in a memory that the processor can reference, and at least one of the intermediate representation and the preprocessing result may be configured to be stored or provided for storage, transmission or logging.

[0047] In a storage medium according to an embodiment of the present invention for solving the above problem, the preprocessing result may be configured to be stored, transmitted, or logged in a form that does not include the input to be processed and consists of at least one of initial condition information, constraints, or candidate restriction information.

[0048] In a storage medium according to an embodiment of the present invention for solving the above problem, the program instruction may be configured to be executed in at least one of an on-device environment or a server environment.

[0049] In a storage medium according to an embodiment of the present invention for solving the above problem, at least a portion of the program instructions may be configured to be executed by hardware accelerator logic or to include instructions in a form that can be referenced or called by hardware accelerator logic.

[0050] In a storage medium according to an embodiment of the present invention for solving the above problem, during the execution process of the program instruction, a language-specific parameter set including a language-specific unit inventory and transition and combination constraints may be configured to be stored or loaded in a memory that the processor can reference and applied to the generation of the preprocessing result.

[0051] In a storage medium according to an embodiment of the present invention for solving the above problem, in the process of providing the preprocessing result to subsequent processing, a data structure, metadata record, or log record including at least one of the initial condition information or the constraint or candidate restriction information may be generated and transmitted to subsequent processing, or configured to be stored, transmitted, or logged. Effects of the invention

[0052] According to an embodiment of the present invention, the computational burden can be alleviated by contributing to limiting the candidate and search space to be considered in subsequent processing, and the stability of the result can be improved by lowering the variability of the processing start state by explicitly setting initial conditions, and it can help alleviate processing delays and improve resource efficiency by reducing unnecessary searches and iterations.

[0053] According to an embodiment of the present invention, it can contribute to reducing overall performance degradation by mitigating the extent to which judgment errors in the preceding stage are propagated to subsequent stages. Furthermore, it can respond to changes in the operating environment or data characteristics by adjusting the initial condition setting and constraint provision methods, which may be advantageous for expanding the scope of application and maintenance. Additionally, differences between languages ​​can be separated into language-specific parameter sets, allowing for expansion without changing the core procedure. Moreover, it can improve the reliability of field voice-based interfaces by mitigating misrecognition and error propagation under industrial site conditions such as high noise, reverberation, and multiple speakers, and by reducing unnecessary searching and repetition, thereby contributing to a reduction in processing delays and the frequency of reprocessing. For example, if initial condition settings and candidate restrictions are configured to be applied to safety-related voices (alarms, calls, instructions, acknowledgments, etc.), it can contribute to lowering the risk of malfunction in safety operations (but is not limited thereto).

[0054] The effects of the present invention are not limited to those mentioned above, and other unmentioned effects will be clearly understood by those skilled in the art from the description in the claims. Brief explanation of the drawing

[0055] FIG. 1 is a diagram illustrating the overall configuration of an initial condition setting system for voice processing according to an embodiment of the present invention. FIG. 2 is a flowchart illustrating the overall flow of a method for setting initial conditions for voice processing according to an embodiment of the present invention. FIG. 3 is a flowchart illustrating a step-by-step processing procedure based on a VC anchor when applying a standard procedure loop according to an embodiment of the present invention. Figure 4 is a table showing the conditions and consequences of a standard procedure loop according to an embodiment of the present invention. FIG. 5 is a flowchart illustrating the processing flow when applying an exception path according to an embodiment of the present invention. Figure 6 is a table showing the mapping relationship between the representative final consonant groups according to an embodiment of the present invention. FIG. 7 is a diagram illustrating the concept of applying a language-specific parameter set according to an embodiment of the present invention. Specific details for implementing the invention

[0056] The present invention is not limited to the embodiments described below but can be implemented in various different forms. These embodiments are merely illustrative of the content of the invention and are provided to inform those skilled in the art of the scope of the invention in detail. The present invention is defined only by the scope of the claims. Throughout the specification, the same reference numerals refer to the same components.

[0057] The embodiments described herein will be explained with reference to the drawings, which are exemplary illustrations of the invention. In the drawings, illustrated areas are depicted for the effective explanation of the technical content. Although terms such as first, second, third, etc., have been used to describe various components in the various embodiments of this specification, these components should not be limited by such terms. These terms are used merely to distinguish one component from another.

[0058] The terms used herein are for describing embodiments and are not intended to limit the invention. In this specification, the singular form includes the plural form unless specifically stated otherwise in the text. As used herein, "comprising" and / or "comprising" does not exclude the presence or addition of one or more other components, steps, actions, and / or elements to the mentioned components, steps, actions, and / or elements.

[0059] Unless otherwise defined, all terms used in this specification may be used in a meaning that is commonly understood by those skilled in the art to which the present invention pertains. Additionally, terms defined in commonly used dictionaries are not to be interpreted ideally or excessively unless explicitly and specifically defined otherwise.

[0060] Hereinafter, the concept of the present invention and embodiments according thereto will be described in detail with reference to the drawings.

[0061] FIG. 1 illustrates an initial condition setting system for voice processing according to an embodiment of the present invention.

[0062] According to one embodiment, an initial condition setting system (1000) for voice processing may include at least one of an input receiving unit (100), a judgment unit (200), an initial condition information generating unit (300), an initial state setting unit (400), a constraint or candidate restriction information generating and providing unit (500), and an output unit or providing unit (600).

[0063] According to one embodiment, an input receiving unit (100) may be configured to receive a processing target input comprising at least one of a voice signal or a unit sequence derived from voice. The processing target input may be included together in a combined form of a voice signal and a unit sequence provided to correspond to the voice signal, and may further include information regarding processing intervals and boundaries. The voice signal may be received in real-time or non-real-time and may be provided in various forms, such as file-based input, streaming input, or frame-based input.

[0064] According to one embodiment, the determination unit (200) may be configured to determine at least one of a target processing section, a boundary, and a processing target unit for the input, and to determine whether to apply a standard procedure loop and whether to apply an exception path. The determination unit (200) may determine the start and end points of the target processing section from the input and determine the boundary of the processing target unit within the target processing section. Additionally, the determination unit (200) may determine whether to apply a standard procedure loop or an exception path based on the determination result, and the determination result may be used to determine a processing path for a subsequent step.

[0065] According to one embodiment, the initial condition information generation unit (300) may be configured to generate initial condition information based on the input and judgment results. The initial condition information is shear initialization information generated based on the input and judgment results and may include VC anchor and / or preliminary articulation setting information, boundary and transition judgment assistance information, and a state value or metadata that can be referenced in subsequent processing. The initial condition information generation unit (300) may generate VC anchor information when a standard procedure loop is applied, and may generate information for exception state processing when an exception path is applied.

[0066] According to one embodiment, the initial state setting unit (400) may be configured to set the VC anchor, which is an initial state reference point, to an initial state based on the initial condition information, or to set the VC anchor setting to conditionally bypass the setting, depending on the determination result regarding whether the standard procedure loop is applied and whether the exception path is applied. If the application of the standard procedure loop is determined, the initial state setting unit (400) may set the VC anchor to an initial state according to the initial condition information and, if necessary, set preliminary articulation information coupled to the VC anchor. If the application of the exception path is determined, the initial state setting unit (400) may conditionally bypass the VC anchor setting and generate record or display information reflecting the exception state.

[0067] According to one embodiment, the constraint or candidate restriction information generation and provision unit (500) generates, sets, or updates constraint or candidate restriction information for restricting search candidates in subsequent processing, and may be configured so that said constraint or candidate restriction information is provided to subsequent processing or referenced and applied during the search process of subsequent processing. The constraint or candidate restriction information may include a list of allowed or prohibited candidates, a state transition restriction mask, a set of candidate indices, or metadata regarding candidate priority or weight, score bias or penalty, or control information regarding thresholds. The constraint or candidate restriction information may be configured to restrict the activation of search candidates or state transition paths in subsequent processing.

[0068] According to one embodiment, an output unit or a providing unit (600) may be configured to provide a preprocessing result including the initial condition information and the constraint or candidate restriction information to subsequent processing. The output unit or a providing unit (600) may transmit the preprocessing result to a subsequent processing module, perform output for transmission, or provide it in a state that can be referenced through memory or an interface. The preprocessing result may include initial condition information, constraint or candidate restriction information, and record or display information regarding the same, and may further include metadata records or log records depending on the implementation.

[0069] According to various embodiments, the system (1000) may further include hardware accelerator logic (700). The hardware accelerator logic (700) may be configured to accelerate the processing of at least some of the determination unit (200), the initial condition information generation unit (300), and the constraint or candidate restriction information generation and provision unit (500), and may improve computational efficiency in terms of delay, power consumption, throughput and / or memory access amount or bandwidth.

[0070] According to various embodiments, at least some of the components of the system (1000) may be placed in an on-device environment, and the remaining parts may be placed in a server environment to communicate with each other and perform distributed operations. For example, parts of the input receiving unit (100) and the determination unit (200) may be placed in an on-device environment, and parts of the constraint or candidate restriction information generation and provision unit (500) may be placed in a server environment, and the on-device environment and the server environment may communicate with each other by transmitting the preprocessing result or metadata records regarding the preprocessing result. For example, it may determine whether voice input from a smartphone / wearable / vehicle, equipment-mounted terminal, etc., includes safety-related commands or acknowledgments, and may provide initial conditions to limit the candidates to be searched by the subsequent recognition step according to the determination result. Additionally, it may be applied for the purpose of reducing the risk caused by misrecognition in voice interfaces directly related to safety, such as equipment control, work instruction confirmation, and danger alarm propagation (but not limited thereto).

[0072] FIG. 2 is a flowchart illustrating the overall flow of a method for setting initial conditions for voice processing according to an embodiment of the present invention.

[0073] According to one embodiment, a method for setting initial conditions for voice processing may be composed of an input receiving step (S100), a judgment step (S200), an initial condition calculation step (S300), an initial state setting step (S400), a constraint providing step (S500), and an output step (S600).

[0074] In step S100, a processing target input may be received that includes at least one of a voice signal or a unit sequence derived from voice. The processing target input may include the voice signal and the unit sequence provided to correspond to the voice signal in a combined form.

[0075] In step S200, at least one of a target processing section, a boundary, and a processing target unit is determined for the input, and whether to apply a standard procedure loop or an exception path can be determined. In the determination step, the start and end points of the target processing section are determined from the input, and the boundary of the processing target unit within the target processing section is determined; and based on the determination result, it is determined whether to apply a standard procedure loop or an exception path.

[0076] In step S300, initial condition information can be calculated based on the above input and judgment results. The initial condition information may include VC anchor and / or preliminary articulation setting information, boundary and transition judgment assistance information, and status values ​​or metadata that can be referenced in subsequent processing.

[0077] In step S400, depending on the determination result regarding whether the standard procedure loop is applied and whether the exception path is applied, the VC anchor, which is the initial state reference point, may be set to an initial state based on the initial condition information, or the VC anchor setting may be set to be conditionally bypassed. If the application of the standard procedure loop is determined, the VC anchor may be set to an initial state according to the initial condition information, and preliminary articulation information coupled to the VC anchor may be set if necessary. If the application of the exception path is determined, the VC anchor setting may be conditionally bypassed, and record or display information reflecting the exception state may be generated.

[0078] In step S500, constraint or candidate restriction information for limiting search candidates for subsequent processing may be created, set, or updated, and said constraint or candidate restriction information may be provided to the subsequent processing or referenced and applied during the search process of the subsequent processing. said constraint or candidate restriction information may include a list of allowed or prohibited candidates, a state transition restriction mask, a set of candidate indices, metadata regarding candidate priority or weight, score bias or penalty, control information regarding thresholds, etc.

[0079] In step S600, a preprocessing result including the initial condition information and the constraint or candidate restriction information may be provided to subsequent processing. The preprocessing result may be transmitted to a subsequent processing module or provided in a referenceable state through memory or an interface.

[0080] According to one embodiment, at least some of steps S200 to S500 may be performed by hardware accelerator logic, in which case computational efficiency in terms of delay, power consumption, throughput and / or memory access amount or bandwidth may be improved.

[0081] According to one embodiment, at least some of steps S200 to S500 may be performed in an on-device environment, and the remaining parts may be performed in a server environment, and the on-device environment and the server environment may communicate with each other by transmitting the preprocessing result or metadata records regarding the preprocessing result.

[0083] FIG. 3 is a flowchart illustrating a step-by-step processing procedure based on a VC anchor when applying a standard procedure loop according to an embodiment of the present invention.

[0084] According to one embodiment, when a standard procedure loop is applied, the processing after step S400 of FIG. 2 may be specified as a VC anchor setting step (S410), a consonant type determination step (S420), a vowel type determination step (S430), and a termination type determination step (S440). For convenience of explanation, the operation of setting the initial state reference point (VC anchor) performed in the initial state setting step (S400) and / or the VC anchor setting step (S410) may be collectively referred to as "VC anchor setting." However, this is merely a name for illustrative purposes, and the specific configuration, sequence, and scope of inclusion of the VC anchor setting may vary depending on the implementation (but are not limited thereto).

[0085] In step S410, a VC anchor can be set to an initial state for setting initial conditions for a step prior to subsequent processing based on input and judgment results. The VC anchor refers to an initial state set for setting initial conditions for a step prior to subsequent processing at the starting point of the target processing interval of a speech signal or unit sequence, and provides initial conditions for determining consonant type, vowel type, and termination as a reference state referenced by subsequent steps. The VC anchor may be configured to include a specific point in time or interval on the time axis, or a state value or metadata representing it. If necessary, preliminary articulation information coupled to the VC anchor may be calculated or set, and the preliminary articulation information refers to a stabilization parameter to assist in branch stabilization and boundary or transition judgment in subsequent steps.

[0086] In step S420, based on the input and judgment results, the type of the consonant stage is C or C t It can be decided as one of the above C. trepresents the reinforcement state of a consonant, and in some implementations, a specific friction reinforcement series may be separated and processed as a non-anchor-type exception path. If the above VC anchor is set, the consonant type can be determined by referencing the above VC anchor. Additionally, separate from the result of determining the consonant type, record or display information indicating the surface realization of the consonant stage may be generated, and said record or display information may be included in the preprocessing result in various forms such as frames, tokens, labels, and status sequences, and provided for subsequent processing (but is not limited thereto). The above C / C t The determination can be made based on observable features or model outputs such as closure or explosion characteristics, tension, energy change, and spectral cues, and if preliminary articulation information is set, the articulation intensity of the subsequent consonant can be determined by referring to the preliminary articulation information.

[0087] In step S430, the vowel structure following the consonant stage can be determined as one of V / IV / VV. The determination of V / IV / VV can be performed based on transition patterns, spectral or formant changes, duration, or branching rules of a language-specific parameter set. If the vowel type results in VV, a VVSubpos is generated to indicate the progression location of the internal VV transition; however, this is managed only as record or indication information and is not a notation for dividing procedural steps.

[0088] In step S440, termination may result in V or VC′. If termination results in VC′, record or indication information regarding the termination marker (VC′) may be generated or provided to indicate termination. The termination marker (VC′) is a termination marker notation with a prime added to distinguish it as a termination phase, and is not interpreted as presupposing the existence of consonant components or being limited to specific consonants; rather, it functions as a marker to logically distinguish the arrival of a termination phase for subsequent processing regardless of the existence of syllable-final components, etc., in the actual utterance. The termination selection of V / VC′ may be determined based on whether the start condition of the next processing unit is confirmed, boundary reliability, or candidate restriction rules.

[0089] According to one embodiment, the combination of consonant type, vowel type, and termination type determined through steps S410 to S440 may result in one of a finite set of loops, and as an example, a total of 12 standard loops may be formed. For convenience of explanation, the finite set of loops (e.g., a total of 12 standard loops) may be referred to as "12 standard loops" below. The number of loops, whether they are split or merged, their names, and internal combinations may be changed according to input conditions, language-specific parameter sets, or implementation policies (but are not limited thereto). The 12 standard loops are for convenience of explanation, and the combination, splitting, substitution, or change in the number of loops is possible and does not limit the scope of the present invention.

[0090] Figure 4 is a table showing the conditions and consequences of a standard procedure loop according to an embodiment of the present invention.

[0091] According to one embodiment, a set of VC anchor-based standard procedure loops can be configured to result in one of a finite number of standard loops depending on a combination of consonant type, vowel type, and termination type. Referring to FIG. 4, the resulting loop is a consonant type (C or C tIt can be determined by a combination of ), vowel type (V, IV, or VV), and ending type (V or VC′).

[0092] For example, if the consonant type is determined to be C, the vowel type is determined to be V, and the ending type is determined to be V, it can result in loop L1. If the consonant type is determined to be C, the vowel type is determined to be V, and the ending type is determined to be VC′, it can result in loop L2. In this way, depending on the combination of consonant type, vowel type, and ending type, it can result in one of 12 standard loops (L1 to L12).

[0093] The above loop ID is not a name of a procedure step, but an identifier to distinguish the flow to which the combination of processing results of the above steps superficially result. When the collection type results in VV, the termination type is determined to be either V or VC′, and when the termination type is determined to be V, the internal realization coordinates of that VV can be indicated by record or display information such as VVSubpos.

[0095] FIG. 5 is a flowchart illustrating the processing flow when applying an exception path according to an embodiment of the present invention.

[0096] According to one embodiment, if the application of an exception path is determined at step S200 of FIG. 2, exception path processing different from the standard procedure loop may be performed. An exception path refers to a branch path configured to conditionally bypass VC anchor settings when it is determined not to apply the standard procedure loop, and to provide record or display information reflecting the exception state and / or constraint or candidate restriction information for subsequent processing.

[0097] According to one embodiment, if the initial consonant of the target processing unit is 'ㅇ' or 'ㅎ', it may branch into a first exception path (immediate vowel entry type) instead of a standard loop set. In this case, it may be configured to transition to a vowel type determination step after reflecting the first exception branching state in the consonant type determination step without generating a VC anchor. In the case of 'ㅎ', even if a glottal fricative or aspirated component is realized simultaneously with the vowel entry, it may be treated as being realized as an initial consonant component within the first exception path rather than as a VC anchor. Record or display information indicating whether the first exception branch has occurred and / or constraint or candidate restriction information to be provided for subsequent processing may be generated and included in the preprocessing result or provided for subsequent processing, and the subsequent termination is maintained to result in either V or VC′. The record or display information generated in relation to the exception branch is not limited to a simple status log or output, but may include information provided to be utilized for limiting search candidates, setting priorities, or adjusting control values ​​in subsequent speech processing or recognition steps (but is not limited thereto).

[0098] According to one embodiment, if the initial consonant of the target processing unit is 'ㅅ' or 'ㅆ', it may branch into a second exception path (frictional leading type) outside the standard loop set. In this case, 'ㅅ' / 'ㅆ' is a non-anchor type fricative group that does not generate a VC anchor, and may be configured to reflect the second exception branching state in the consonant type determination step, form a leading fricative phase once, and then transition to the vowel type determination step. The leading fricative phase does not refer to a VC anchor or a separate new anchor. Record or display information indicating whether the second exception branching occurs and whether the leading fricative phase is formed, and / or constraint or candidate restriction information to be provided for subsequent processing, may be generated and included in the preprocessing result or provided for subsequent processing, and the subsequent termination is maintained to result in either V or VC′. The record or display information generated in relation to the exception branching is not limited to a simple status log or output, but may include (but is not limited to) information provided to be utilized for limiting search candidates, setting priorities, or adjusting control values ​​in subsequent speech processing or recognition steps.

[0100] Figure 6 is a table showing the mapping relationship between the representative final consonant groups according to an embodiment of the present invention.

[0101] According to one embodiment, by mapping a group of final consonants to a predefined representative symbol, the internal notation used for generating VC anchors and, if necessary, the representation of record or display information can be simplified, and the configuration of constraint or candidate restriction information provided for subsequent processing can be assisted. Referring to FIG. 6, an input group of final consonants can be mapped to a representative symbol, and the representative symbol generated by the mapping can be included in the preprocessing result as a representative result in which a processing unit corresponding to a syllable-final component is mapped to a representative value.

[0102] For example, the ㄱ series (ㄱ / ㄲ / ㅋ) can be mapped to the representative symbol ㄱ, the ㄷ series (ㄷ / ㄸ / ㅈ / ㅉ / ㅊ / ㅌ) can be mapped to the representative symbol ㅅ, and the ㅂ series (ㅂ / ㅃ / ㅍ) can be mapped to the representative symbol ㅂ. Additionally, the ㄴ series (ㄴ) can be mapped to the representative symbol ㄴ, the ㄹ series (ㄹ) to the representative symbol ㄹ, and the ㅁ series (ㅁ) to the representative symbol ㅁ. The above mappings and representative symbols are examples, and the mapping is not dependent on the specific character itself but can be performed by predefined representative symbol mapping rules, and can be replaced or modified in the form of rules, tables, or functions, without limiting the scope of the present invention.

[0103] According to one embodiment, non-anchor type initial consonants (ㅇ, ㅎ, and ㅅ / ㅆ) are exceptions to VC anchor generation and may be excluded from the application of this representative mapping. Representativeization is an internal notation or internal mapping for VC anchor generation and, if necessary, simplification of recording or display, and the identity and consonant type of the target initial consonant may be maintained and confirmed in the consonant type determination step. Additionally, since the representativeization value may be shared by multiple consonant groups, the target consonant is not interpreted as being fully identified solely by the representativeization step.

[0104] Here, the representative symbol (e.g., 'ㅅ') refers to an internal notation or internal mapping based on the representative mapping of the final consonant group, and should not be interpreted as corresponding one-to-one with or directly referring to the initial consonant or surface realization of the actual input speech.

[0105] Therefore, the determination of exception paths related to the initial consonant 'ㅅ / ㅆ' (see Fig. 5) is not determined by the representative symbol itself, but can be performed independently based on input characteristics and the result of the determination (not limited thereto).

[0107] FIG. 7 is a diagram illustrating the concept of applying a language-specific parameter set according to an embodiment of the present invention.

[0108] According to one embodiment, the core procedure of the present invention is not limited to a specific language or a specific notation system, and language-specific differences may be applied by separating them into language-specific parameter sets including language-specific unit inventories and transition and combination constraints. Referring to FIG. 7, the language-general core procedure is maintained invariantly and may be extended by adding or modifying language-specific parameter sets.

[0109] The above language-specific parameter set may be configured to include classification criteria and branching rules for consonants, vowels, and finals, as well as combination constraints and exception paths. Depending on the language, permissible characteristics of consonant initiation, plural combinations, types of nuclei, prosodic markers, etc., may be applied separately as distinct parameters or additional processing. For example, if the permissible range of plural combinations varies by language, the permissible range and transition and combination constraints may be included in the language-specific parameter set and applied to the core procedure. Additionally, if prosodic markers are involved, such markers may be configured as language-specific parameters separated from the core procedure or as separate additional processing and applied selectively.

[0110] According to one embodiment, language extension can be achieved by adding or modifying the parameter set without changing the core structure, and accordingly, the present invention provides a language-general structure. The language-specific parameter set may be configured to be stored or loaded in a storage medium or memory and applied to the generation of a preprocessing result.

[0111] As such, various embodiments of the present invention describe a system and method for setting initial conditions for voice processing. The record or display information is not limited to information for simply outputting the processing results of the present invention to the outside, but may be provided as input information for controlling the search space in a subsequent voice processing or recognition step (though not limited thereto). By utilizing the system and method described in the various embodiments of the present invention, the computational burden can be alleviated by limiting the candidate and search spaces to be considered in subsequent processing, the stability of results can be improved by reducing the variability of the processing start state by explicitly setting initial conditions, and processing delays and resource efficiency can be improved by reducing unnecessary searches and repetitions.

[0112] Meanwhile, although it has been described that all components constituting an embodiment of the present invention are combined or operate in combination, the present invention is not necessarily limited to such an embodiment. That is, within the scope of the purpose of the present invention, all components may be selectively combined in one or more ways to operate.

[0113] In addition, while all such components may each be implemented as a single independent piece of hardware, some or all of the components may be optionally combined to be implemented as a computer program having a program module that performs some or all of the combined functions on one or more pieces of hardware. The codes and code segments constituting the computer program will be readily deducible by those skilled in the art of the present invention.

[0114] An embodiment of the present invention can be implemented by storing such computer programs on a non-transient storage medium that is readable by a computer, and by reading and executing them by a computer. Here, a non-transient readable recording medium refers to a medium that stores data semi-permanently and is readable by a device, rather than a medium that stores data for a short period of time, such as a register, cache, or memory. Specifically, the aforementioned programs may be provided by being stored on a non-transient readable recording medium such as a CD, DVD, hard disk, Blu-ray disc, USB, memory card, ROM, etc.

[0115] Although preferred embodiments of the present invention have been illustrated and described above, the present invention is not limited to the specific embodiments described above. It is understood that various modifications can be made by those skilled in the art without departing from the essence of the invention as claimed in the claims, and such modifications should not be understood individually from the technical spirit or perspective of the present invention. Explanation of the symbols

[0116] 1000: Initial condition setting system for voice processing 100: Input receiver 200: Judgment Division 300: Initial condition information generation section 400: Initial state setting section 500: Restriction or Candidate Restriction Information Generation and Provision Unit 600: Output section or providing section 700: Hardware Accelerator Logic

Claims

Claim 1 A method performed by an initial condition setting system for voice processing, comprising: (a) receiving a processing target input including at least one of a voice signal or a unit sequence derived from voice; (b) determining at least one of a target processing interval, a boundary, and a processing target unit for the input, and determining whether to apply a standard procedure loop and whether to apply an exception path; (c) calculating initial condition information based on the input and the determination result; (d) setting a VC anchor, which is an initial state reference point, to an initial state based on the initial condition information according to the determination result regarding whether to apply a standard procedure loop and whether to apply an exception path, or setting the VC anchor setting to conditionally bypass; (e) generating, setting, or updating constraint or candidate restriction information to limit search candidates for subsequent processing, and providing the constraint or candidate restriction information to subsequent processing or allowing it to be referenced and applied during the search process of subsequent processing; and (f) providing a preprocessing result including the initial condition information and the constraint or candidate restriction information to subsequent processing. Claim 2 In claim 1, when the application of the standard procedure loop is determined, step (d) comprises: (d-1) setting the VC anchor to an initial state; (d-2) the consonant type to C or C t A method configured to include: (d-3) a step of determining one of the vowel types as V, IV, or VV; and (d-4) a step of processing the ending as V or VC′. Claim 3 A method according to claim 2, wherein the combined result of steps (d-2) through (d-4) is configured to result in one of a finite set of standard loops predefined according to a combination of consonant type, vowel type, and termination type. Claim 4 A method according to claim 1, wherein when the application of the exception path is determined, the step (d) is configured to conditionally bypass the VC anchor setting and generate record or display information reflecting the exception state, and then transition to subsequent processing. Claim 5 A method according to claim 2, wherein the termination process of step (d-4) results in either a vowel termination (V) or a termination marker (VC′), and the termination marker (VC′) is not limited to a specific consonant component or presupposes the existence of a specific consonant component, and is configured to function as a marker for logically distinguishing the arrival of a termination phase. Claim 6 A method according to claim 1, wherein the constraint or candidate restriction information comprises at least one of an allowed candidate or prohibited candidate list, a state transition restriction mask, a set of candidate indices, metadata regarding candidate priority or weight, a score bias or penalty, or control information regarding a threshold, and configured to restrict the activation of a search candidate or state transition path in subsequent processing. Claim 7 A method according to claim 1, wherein at least some of the steps (b), (c), and (e) are performed by hardware accelerator logic and configured to improve delay, power consumption, throughput, or memory access efficiency. Claim 8 A method according to claim 2, wherein the preprocessing result provided in step (f) is configured to include record or display information indicating the surface realization of the consonant step determined in step (d-2). Claim 9 A method according to claim 2, wherein if the vowel type is determined to be VV in step (d-3), the preprocessing result provided in step (f) is configured to include VVSubpos record or display information indicating the progression position from the preceding vowel component to the succeeding vowel component. Claim 10 A method according to claim 1, wherein step (c) includes a step of mapping a processing unit corresponding to a syllable-ending component to a representative value according to a predefined mapping rule, and the preprocessing result provided in step (f) is configured to include the representative value mapping result. Claim 11 A method according to claim 1, wherein language-specific differences are separated and applied as a language-specific parameter set including a language-specific unit inventory and transition and combination constraints, and configured to be expandable without changing steps (a) through (f) by adding or modifying the language-specific parameter set. Claim 12 A method according to claim 1, wherein at least one of step (b) and step (c) is configured to be performed by rule-based processing, statistical model-based processing, or a combination thereof. Claim 13 A method according to claim 12, wherein at least one of step (b) and step (c) is configured to dynamically select and perform at least one of rule-based processing and model-based processing according to input characteristics or execution environment. Claim 14 A method according to claim 2, wherein the initial condition information calculated in step (c) includes preliminary articulation setting information that affects the articulation intensity of a subsequent consonant, and step (d-1) is configured to set a VC anchor by reflecting the preliminary articulation setting information. Claim 15 A method according to claim 1, wherein at least some of the steps (b), (c), and (e) are performed in an on-device environment and the remainder are performed in a server environment, and the on-device environment and the server environment are configured to communicate with each other by transmitting the preprocessing result or metadata records regarding the same. Claim 16 A method according to claim 1, wherein the step (c) for the next processing unit is configured to be performed by referring to at least a portion of the preprocessing results of the step (f) performed for the immediate processing unit. Claim 17 A system for setting initial conditions for voice processing, comprising: an input receiving unit configured to receive a processing target input including at least one of a voice signal or a unit sequence derived from voice; a determination unit configured to determine at least one of a target processing interval, a boundary, and a processing target unit for the input, and to determine whether to apply a standard procedure loop and whether to apply an exception path; an initial condition information generating unit configured to calculate initial condition information based on the input and the determination result; an initial state setting unit configured to set a VC anchor, which is an initial state reference point, to an initial state based on the initial condition information, or to set the VC anchor setting to bypass conditionally, according to the determination result regarding whether to apply a standard procedure loop and whether to apply an exception path; a constraint or candidate restriction information generating and providing unit configured to generate, set, or update constraint or candidate restriction information for limiting search candidates for subsequent processing, and such constraint or candidate restriction information is provided to subsequent processing or referenced and applied during the search process of subsequent processing; and an output unit or providing unit configured to provide a preprocessing result including the initial condition information and the constraint or candidate restriction information to subsequent processing. Claim 18 A system according to claim 17, further comprising hardware accelerator logic configured to accelerate the processing of at least some of the determination unit, the initial condition information generation unit, and the constraint or candidate restriction information generation and provision unit. Claim 19 In claim 17, when the application of the standard procedure loop is determined, the initial state setting unit sets the VC anchor, determines the consonant type (C or C t A system configured to sequentially perform ), determine vowel type (V, IV, or VV), and terminate process (V or VC′). Claim 20 A system according to claim 19, wherein the initial state setting unit is configured such that a combination of consonant type, vowel type, and termination type results in one of a predefined finite set of standard loops. Claim 21 A system according to claim 17, wherein when the application of the exception path is determined, the initial state setting unit is configured to conditionally bypass the VC anchor setting and generate record or display information reflecting the exception state to transition to subsequent processing. Claim 22 A system according to claim 19, wherein the initial state setting unit determines the termination process to either a vowel termination (V) or a termination by a termination marker (VC′), and the termination marker (VC′) is configured to function as a marker that logically distinguishes the arrival of a termination phase, without presupposing the existence of a specific consonant component or being limited to a specific consonant. Claim 23 A system according to claim 17, wherein the constraint or candidate restriction information generation and provision unit is configured to generate constraint information comprising at least one of an allowed candidate or prohibited candidate list, a state transition restriction mask, a set of candidate indices, metadata regarding candidate priority or weight, score bias or penalty, or control information regarding threshold, to restrict a search candidate or state transition path for subsequent processing. Claim 24 A system according to claim 19, wherein the output or providing unit is configured to include record or display information indicating a surface realization of a consonant stage in the preprocessing result. Claim 25 A system according to claim 19, wherein when the vowel type is determined to be VV, the output unit or providing unit is configured to include VVSubpos record or display information indicating the progression position from a preceding vowel component to a succeeding vowel component in the preprocessing result. Claim 26 A system according to claim 17, wherein the initial condition information generating unit maps a processing unit corresponding to a syllable-ending component to a representative value according to a predefined mapping rule, and the output unit or providing unit is configured to include the representative value mapping result in the preprocessing result. Claim 27 A system according to claim 17, wherein at least one of the determination unit and the constraint or candidate restriction information generation and provision unit is configured to generate different determination and constraint information for each language by applying a language-specific parameter set including a language-specific unit inventory and transition and combination constraint conditions. Claim 28 A system according to claim 17, wherein at least one of the determination unit and the initial condition information generation unit is configured to operate by rule-based processing, statistical model-based processing, or a combination thereof. Claim 29 A system according to claim 28, wherein at least one of the determination unit and the initial condition information generation unit is configured to dynamically select and perform at least one of rule-based processing and model-based processing according to input characteristics or execution environment. Claim 30 A system according to claim 19, wherein when the application of the standard procedure loop is determined, the initial condition information generating unit calculates initial condition information including preliminary articulation setting information that affects the articulation intensity of a subsequent consonant, and the initial state setting unit is configured to set a VC anchor by reflecting the preliminary articulation setting information. Claim 31 A system according to claim 17, wherein at least some of the determination unit, the initial condition information generation unit, and the constraint or candidate restriction information generation and provision unit are placed in an on-device environment and the remainder are placed in a server environment, and the on-device environment and the server environment are configured to communicate with each other by transmitting the preprocessing result or metadata records related thereto. Claim 32 A system according to claim 17, wherein the initial condition information generating unit is configured to refer to at least a portion of the preprocessing result provided for the immediate processing unit when calculating initial condition information for the next processing unit. Claim 33 A computer-readable storage medium storing program instructions that, when executed by a processor, cause the method described in claim 1 to be performed. Claim 34 A storage medium according to claim 33, wherein, during the execution of the program instruction, at least one of an intermediate representation including a feature sequence, a status sequence or metadata and the preprocessing result is loaded or updated in memory accessible to the processor, and at least one of the intermediate representation and the preprocessing result is configured to be stored, transmitted, or logged. Claim 35 A storage medium according to claim 33, wherein the preprocessing result does not include the input to be processed and is configured to be stored, transmitted, or logged in a form consisting of at least one of initial condition information, constraints, or candidate restriction information. Claim 36 A storage medium according to claim 33, wherein the program instruction is configured to be executed in at least one of an on-device environment or a server environment. Claim 37 A storage medium according to claim 33, wherein at least a portion of the program instructions is configured to include instructions that are executed by hardware accelerator logic or call hardware accelerator logic. Claim 38 A storage medium according to claim 33, wherein, during the execution of the program instruction, a language-specific parameter set including a language-specific unit inventory and transition and combination constraints is loaded into a memory accessible to the processor and configured to be applied to the generation of the preprocessing result. Claim 39 A storage medium according to claim 33, wherein, in the process of providing the preprocessing result to subsequent processing, a data structure, metadata record, or log record comprising at least one of the initial condition information or the constraint or candidate restriction information is generated and delivered to subsequent processing, or is configured to be stored, transmitted, or logged.

Images