Voice Processing Method, Device, Electronic Device and Medium

Abstract

The present disclosure relates to a voice processing method, apparatus, electronic device and medium, and particularly relates to the technical field of voice processing. Among them, the method includes: recognizing voice data to obtain corresponding target recognition text and target voiceprint features; determining a target user according to the target voiceprint features; determining target information corresponding to the target recognition text based on an error correction map corresponding to the target user, where the error correction map includes: the correspondence between the target recognition text and the target information; obtaining the similarity between the target recognition text and the target information, and if the similarity exceeds a preset threshold, modifying the target recognition text to the target information. The embodiments of the present disclosure can correct the voice data of the target user, which is beneficial to improving the error correction speed and the user experience.

Description

Technical Field

[0001] The present disclosure relates to the technical field of speech processing, and in particular, to a speech processing method, apparatus, electronic device, and medium. Background Art

[0002] With the popularization of voice assistants, more and more electronic devices, such as various household appliances and terminal devices, have voice recognition functions, which facilitate users to search and control through voice. When voice data is incorrect, it will directly affect the search results and control results. Therefore, speech processing becomes particularly important.

[0003] Speech processing can be targeted at different groups of people. For example, in recent years, preschool children have accounted for an increasing proportion of the population using voice to search and control corresponding electronic devices, and the search volume of children's data in the search data is relatively large. Preschool children have the following problems in voice search and control: First, their statements are relatively subjective, they like to use simple words to describe search content, and they confuse the title and character statements, etc.; Second, they are prone to misidentifying, and they like to create new words. Using traditional speech processing methods, such as edit distance error correction and pinyin similarity error correction, etc., to perform speech processing on preschool children has low accuracy, resulting in difficulty in searching for correct results or determining the true control intention. Summary of the Invention

[0004] In order to solve the above technical problems or at least partially solve the above technical problems, the present disclosure provides a speech processing method, apparatus, electronic device, and medium, which can correct the speech data of the target user, which is beneficial to improving the error correction speed and the user experience.

[0005] In order to achieve the above object, the technical solutions provided by the embodiments of the present disclosure are as follows:

[0006] In a first aspect, the present disclosure provides a speech processing method, the method comprising:

[0007] Identifying the speech data to obtain a corresponding target recognition text and a target voiceprint feature;

[0008] Determining a target user according to the target voiceprint feature;

[0009] Based on the error correction map corresponding to the target user, determining the target information corresponding to the target recognition text, where the error correction map includes: the correspondence between the target recognition text and the target information;

[0010] Obtaining the similarity between the target recognition text and the target information, and if the similarity exceeds a preset threshold, modifying the target recognition text to the target information.

[0011] As an optional implementation manner of an embodiment of the present disclosure, obtaining the similarity between the target recognition text and the target information includes:

[0012] Based on the error correction graph, determine the path information corresponding to the target recognition text;

[0013] According to the path information, determine the similarity between the target recognition text and the target information.

[0014] As an optional implementation manner of an embodiment of the present disclosure, the path information includes: the first path information and the second path information respectively corresponding to the target recognition text under different path types;

[0015] The step of determining the similarity between the target recognition text and the target information according to the path information includes:

[0016] Based on the first path information, determine the first probability value corresponding to the first path type, and based on the second path information, determine the second probability value corresponding to the second path type;

[0017] According to the first probability value and the second probability value, determine the similarity between the target recognition text and the target information.

[0018] As an optional implementation manner of an embodiment of the present disclosure, the path information includes: the relevant probability between the word segmentation represented by each sub-node in the corresponding path of the target recognition text and the target recognition text, and the weight factor corresponding to the word segmentation represented by each sub-node;

[0019] The step of determining the similarity between the target recognition text and the target information according to the path information includes:

[0020] According to the relevant probability and the corresponding weight factor, determine the similarity between the target recognition text and the target information.

[0021] As an optional implementation manner of an embodiment of the present disclosure, the method further includes:

[0022] Obtain the first word segmentation included in the corpus information related to the target information and the second word segmentation included in the label information corresponding to the target information;

[0023] Taking the target information as the central node, the first word segmentation, the second word segmentation, and the generated information as sub-nodes, the first association relationship between the central node and different sub-nodes as the edge between the central node and different sub-nodes, and the second association relationship between each sub-node as the edge between each sub-node, establish the error correction graph.

[0024] As an alternative implementation manner of an embodiment of the present disclosure, the method further includes:

[0025] Determine the probability corresponding to the first word segmentation according to the frequency corresponding to the syntactic dependency relationship of the first word segmentation, or the part-of-speech frequency of the target word segmentation corresponding to the core relationship in the syntactic dependency relationship;

[0026] Determine the probability corresponding to the second word segmentation according to the initial weight and weight factor corresponding to the second word segmentation;

[0027] Based on the probability corresponding to the first word segmentation and the probability corresponding to the second word segmentation, respectively determine the weight values of the corresponding edges in the error correction graph.

[0028] As an alternative implementation manner of an embodiment of the present disclosure, the method further includes:

[0029] If it is determined according to the target voiceprint feature that the user is a non-target user, or the similarity does not exceed the preset threshold, then process the target recognition text to obtain a corresponding processing result;

[0030] Based on the processing result, modify the processing result through a preset speech error correction method to obtain a modified text.

[0031] In a second aspect, the present disclosure provides a voice processing device, and the device includes:

[0032] An identification module, configured to identify voice data to obtain a corresponding target recognition text and a target voiceprint feature;

[0033] A first determination module, configured to determine a target user according to the target voiceprint feature;

[0034] A second determination module, configured to determine target information corresponding to the target recognition text based on an error correction graph corresponding to the target user, where the error correction graph includes: a correspondence between the target recognition text and the target information;

[0035] A modification module, configured to obtain a similarity between the target recognition text and the target information, and if the similarity exceeds a preset threshold, modify the target recognition text to the target information.

[0036] As an alternative implementation manner of an embodiment of the present disclosure, the modification module includes:

[0037] A path information determination unit, configured to determine path information corresponding to the target recognition text based on the error correction graph;

[0038] A similarity determination unit, configured to determine a similarity between the target recognition text and the target information according to the path information;

[0039] A modification unit, configured to modify the target recognition text to the target information if the similarity exceeds a preset threshold.

[0040] As an optional implementation manner of an embodiment of the present disclosure, the path information includes: first path information and second path information respectively corresponding to the target recognition text under different path types;

[0041] The similarity determination unit is configured to:

[0042] Determine a first probability value corresponding to a first path type based on the first path information, and determine a second probability value corresponding to a second path type based on the second path information;

[0043] Determine the similarity between the target recognition text and the target information according to the first probability value and the second probability value.

[0044] As an optional implementation manner of an embodiment of the present disclosure, the path information includes: the correlation probability between the word segmentation represented by each child node of the target recognition text under the corresponding path and the target recognition text, and the weight factor corresponding to the word segmentation represented by each child node;

[0045] The similarity determination unit is further configured to: determine the similarity between the target recognition text and the target information according to the correlation probability and the corresponding weight factor.

[0046] As an optional implementation manner of an embodiment of the present disclosure, the apparatus further includes: an error correction graph building module, configured to:

[0047] Obtain a first word segmentation included in the corpus information related to the target information and a second word segmentation included in the label information corresponding to the target information;

[0048] Use the target information as the central node, the first word segmentation, the second word segmentation, and the generated information as child nodes, use the first association relationship between the central node and different child nodes as the edge between the central node and different child nodes, and use the second association relationship between each child node as the edge between each child node to build the error correction graph.

[0049] As an optional implementation manner of an embodiment of the present disclosure, the apparatus further includes: a weight value determination module, configured to:

[0050] Determine the probability corresponding to the first word segmentation according to the frequency corresponding to the syntactic dependency relationship of the first word segmentation, or the part-of-speech frequency of the target word segmentation corresponding to the core relationship in the syntactic dependency relationship;

[0051] Determine the probability corresponding to the second participle according to the initial weight and the weight factor corresponding to the second participle;

[0052] Based on the probability corresponding to the first participle and the probability corresponding to the second participle, determine the weight values of the corresponding edges in the error correction graph respectively.

[0053] As an optional implementation manner of the embodiments of the present disclosure, the apparatus further includes:

[0054] A processing module, configured to process the target recognition text to obtain a corresponding processing result if it is determined according to the target voiceprint feature that the user is a non-target user, or the similarity does not exceed the preset threshold;

[0055] A text determination module, configured to modify the processing result by a preset speech error correction method based on the processing result to obtain a modified text.

[0056] In a third aspect, the present disclosure further provides an electronic device, including:

[0057] One or more processors;

[0058] A storage device, configured to store one or more programs,

[0059] When the one or more programs are executed by the one or more processors, the one or more processors implement any one of the speech processing methods in the embodiments of the present disclosure.

[0060] In a fourth aspect, the present disclosure further provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements any one of the speech processing methods in the embodiments of the present disclosure.

[0061] The technical solutions provided by the embodiments of the present disclosure have the following advantages compared with the prior art: First, the speech data is recognized to obtain a corresponding target recognition text and a target voiceprint feature, then the target user is determined according to the target voiceprint feature, and then based on the error correction graph corresponding to the target user, the target information corresponding to the target recognition text is determined. The error correction graph includes: the corresponding relationship between the target recognition text and the target information. Finally, the similarity between the target recognition text and the target information is obtained. If the similarity exceeds the preset threshold, the target recognition text is modified to the target information. Through the above method, the speech data of the target user can be corrected, which is beneficial to improving the error correction speed and the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, showing embodiments consistent with the present disclosure, and are used together with the specification to explain the principles of the present disclosure.

[0063] To more clearly illustrate the technical solutions in the embodiments of the present disclosure or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0064] Figure 1 Schematic diagram of an application scenario of a voice processing process in an embodiment of the present disclosure;

[0065] Figure 2A Hardware configuration block diagram of an electronic device according to one or more embodiments of the present disclosure;

[0066] Figure 2B Schematic diagram of the software configuration of an electronic device according to one or more embodiments of the present disclosure;

[0067] Figure 2C Schematic diagram of the display of an icon control interface of an application program included in an intelligent device according to one or more embodiments of the present disclosure;

[0068] Figure 3A Schematic flow diagram of a voice processing method provided by an embodiment of the present disclosure;

[0069] Figure 3B Schematic diagram of the principle of a voice processing method provided by an embodiment of the present disclosure;

[0070] Figure 4A Schematic flow diagram of another voice processing method provided by an embodiment of the present disclosure;

[0071] Figure 4B Schematic diagram of the principle of another voice processing method provided by an embodiment of the present disclosure;

[0072] Figure 5A Schematic flow diagram of a method for establishing an error correction map provided by an embodiment of the present disclosure;

[0073] Figure 5B Schematic diagram of the principle of a method for establishing an error correction map provided by an embodiment of the present disclosure;

[0074] Figure 5C Schematic diagram of a piece of knowledge in the process of establishing an error correction map provided by an embodiment of the present disclosure;

[0075] Figure 5D Schematic diagram of an error correction map provided by an embodiment of the present disclosure;

[0076] Figure 6A Schematic diagram of the principle of determining the weight value of the corresponding edge in an error correction map provided by an embodiment of the present disclosure;

[0077] Figure 6B Schematic diagram of the principle for determining the probability corresponding to the first word segmentation provided by an embodiment of the present disclosure;

[0078] Figure 6C Schematic diagram of a method for determining similarity based on an error correction map provided by an embodiment of the present disclosure;

[0079] Figure 6D Schematic diagram of another method for determining similarity based on an error correction map provided by an embodiment of the present disclosure;

[0080] Figure 7 Schematic flowchart of yet another voice processing method provided by an embodiment of the present disclosure;

[0081] Figure 8A Schematic diagram of the structure of a voice processing device provided by an embodiment of the present disclosure;

[0082] Figure 8B Schematic diagram of the structure of the modification module in the voice processing device provided by an embodiment of the present disclosure;

[0083] Figure 9 Schematic diagram of the structure of an electronic device provided by an embodiment of the present disclosure. Detailed implementation manners

[0084] In order to be able to more clearly understand the above objects, features and advantages of the present disclosure, the solutions of the present disclosure will be further described below. It should be noted that, without conflict, the embodiments of the present disclosure and the features in the embodiments may be combined with each other.

[0085] Many specific details are set forth in the following description in order to fully understand the present disclosure, but the present disclosure may also be implemented in other ways different from those described herein; obviously, the embodiments in the specification are only some embodiments of the present disclosure, rather than all embodiments.

[0086] The terms "first" and "second" etc. in the present disclosure are used to distinguish different objects, rather than to describe a specific order of the objects. For example, the first path information and the second path information etc. are used to distinguish different path information, rather than to describe the specific order of the path information.

[0087] With the continuous development of technology, various electronic products, such as home appliances, are becoming more and more intelligent, bringing great convenience to people's lives. More and more home appliances have voice search and voice control functions, which facilitate users to achieve quick search and rapid control. Home appliances identify the voice data of users to determine their true intentions and execute subsequent control processes based on these true intentions. With the development of technologies related to voice and natural language processing, voice recognition and processing, as a commonly used human-computer interaction technology, have been widely applied to various electronic products and are loved by the majority of users for their natural and convenient interaction methods, gradually becoming the mainstream interaction control method in the era of intelligent products.

[0088] Exemplarily, according to the analysis of the user population on the big data platform, it is found that preschool children (i.e., children) have a relatively large amount of voice search for data of children's media assets (i.e., children's programs, such as cartoons, children's entertainment programs, etc.) when using the voice search function. There are problems such as strong subjectivity in their statements and easy mistakes in program names during voice search. For the above problems, when using traditional edit distance error correction methods or pinyin similarity error correction methods to process voice data, the accuracy of the obtained text data is not high. In the scenario of incorrect children's program names or other errors, accurate error correction cannot be achieved, resulting in a low reach rate of search services.

[0089] As can be seen from the above, the existing voice processing methods have low accuracy. Therefore, a voice processing method with higher accuracy is needed.

[0090] Figure 1 This is a schematic diagram of the application scenario of a voice processing process in an embodiment of the present disclosure. Exemplarily, such as Figure 1As shown, it is assumed that the smart devices in the smart home scenario include smart device 100 (i.e., smart refrigerator), smart device 101 (i.e., smart washing machine), and smart device 102 (i.e., smart display device). When the user wants to perform voice search or voice control through the smart devices in the home scenario, the user can record through the recording application in the terminal device 104 to obtain voice data. Among them, the voice data can be the user's search intention or control intention. The terminal device 104 sends the user's voice data to the server 103 so that the server 103 executes the corresponding voice processing method, that is: identify the voice data to obtain the corresponding target recognition text and target voiceprint feature, determine the target user according to the target voiceprint feature, determine the target information corresponding to the target recognition text based on the error correction map corresponding to the target user, obtain the similarity between the target recognition text and the target information, and if the similarity exceeds the preset threshold, modify the target recognition text to the target information. After the server 103 obtains the target information, it sends the target information to the corresponding smart device so that the smart device executes the corresponding function. Or the user can enter the voice through the local control device 105, such as the recording module in the Internet of Things terminal. Among them, the voice can be the user's search intention or control intention. The local control device 105 executes the above voice processing method and sends the obtained target information to the smart device so that the smart device searches based on the target information. It is also possible to configure a voice processing device in each smart device, and the voice processing device executes the above voice processing method to achieve the purpose of voice search or voice control.

[0091] It should be noted that: the smart device 101 can also be a tablet, a digital cinema system, a video and audio server, etc. Figure 1 This is only an exemplary description, and there is no specific limitation on the types and numbers of smart devices.

[0092] The voice processing method provided by the embodiments of the present disclosure can be implemented based on an electronic device, or a functional module or functional entity in the electronic device.

[0093] Among them, the electronic device can be a personal computer (PC), a server, a mobile phone, a tablet computer, a laptop computer, a mainframe computer, etc., and the embodiments of the present disclosure do not make specific limitations on this.

[0094] Exemplarily, Figure 2A is a hardware configuration block diagram of an electronic device according to one or more embodiments of the present disclosure. As Figure 2AAs shown, the electronic device includes at least one of a tuner-demodulator 210, a communicator 220, a detector 230, an external device interface 240, a controller 250, a display 260, an audio output interface 270, a memory, a power supply, and a user interface 280. Among them, the controller 250 includes a central processing unit, a video processor, an audio processor, a graphics processor, a RAM, a ROM, and first to nth interfaces for input / output. The display 260 can be at least one of a liquid crystal display, an OLED display, a touch display, and a projection display, and can also be a projection device and a projection screen. The tuner-demodulator 210 receives broadcast television signals through wired or wireless reception, and demodulates audio and video signals from multiple wireless or wired broadcast television signals, such as EPG audio and video data signals. The communicator 220 is a component for communicating with external devices or servers according to various communication protocol types. For example, the communicator can include at least one of a Wifi module, a Bluetooth module, a wired Ethernet module, other network communication protocol chips or near-field communication protocol chips, and an infrared receiver. The electronic device can establish the transmission and reception of control signals and data signals with the server 203 or the local control device 205 through the communicator 220. The detector 230 is used to collect signals from the external environment or for external interaction. The controller 250 and the tuner-demodulator 210 can be located in different split devices, that is, the tuner-demodulator 210 can also be in an external device of the main device where the controller 250 is located, such as an external set-top box, etc. The user interface 280 can be used to receive control signals from a control device (such as an infrared remote control, etc.).

[0095] In some embodiments, the controller 250 controls the operation of the electronic device and responds to user operations through various software control programs stored in the memory. The controller 250 controls the overall operation of the electronic device. If the user inputs a user command on the graphical user interface (GUI) displayed on the display 260, the user input interface receives the user input command through the graphical user interface (GUI). Alternatively, if the user inputs a user command by inputting a specific sound or gesture, the user input interface receives the user input command by identifying the sound or gesture through a sensor.

[0096] In some embodiments, the "user interface" is a medium interface for interaction and information exchange between an application or operating system and a user, which realizes the conversion between the internal form of information and the form acceptable to the user. The commonly used form of user interface is the graphical user interface (GUI), which refers to a user interface related to computer operations displayed in a graphical manner. It can be an interface element such as an icon, window, and control displayed on the display screen of an electronic device, where the control can include at least one of the visual interface elements such as icons, buttons, menus, tabs, text boxes, dialog boxes, status bars, navigation bars, widgets, etc.

[0097] Figure 2B FIG. 1 is a schematic diagram of software configuration of an electronic device according to one or more embodiments of the present disclosure, such as Figure 2B As shown in the figure, the system is divided into four layers, from top to bottom: the application layer (referred to as the "application layer"), the application framework layer (referred to as the "framework layer"), the Android runtime (Android runtime) and system library layer (referred to as the "system runtime library layer"), and the kernel layer.

[0098] In some embodiments, at least one application program runs in the application layer. These applications can be window programs, system settings programs, clock programs, etc. that come with the operating system, or applications developed by third-party developers. In specific implementations, the applications in the application layer include but are not limited to the above examples.

[0099] In some embodiments, the system runtime layer provides support for the upper layer, namely the framework layer. When the framework layer is used, the Android operating system will run the C / C++ library contained in the system runtime layer to implement the functions to be implemented by the framework layer.

[0100] In some embodiments, the kernel layer is a layer between hardware and software, and includes at least one of the following drivers: audio driver, display driver, Bluetooth driver, camera driver, WIFI driver, USB driver, HDMI driver, sensor driver (such as fingerprint sensor, temperature sensor, pressure sensor, etc.), and power driver, etc.

[0101] Figure 2C This is a schematic diagram showing an icon control interface of an application included in a smart device (mainly a smart playback device, such as a smart TV, a digital cinema system, or an audio and video server, etc.) according to one or more embodiments of the present disclosure. Figure 2CAs shown in the figure, the application layer contains at least one application that can display corresponding icon controls on the display, such as: live TV application icon control, video-on-demand (VOD) application icon control, media center application icon control, application center icon control, game application icon control, etc. The live TV application can provide live TV through different signal sources. The video-on-demand (VOD) application can provide videos from different storage sources. Different from the live TV application, the video-on-demand provides video displays from certain storage sources. The media center application can provide various playback applications. The application center can provide storage for various applications.

[0102] The voice processing method provided by the embodiment of the present application can be implemented based on the above-mentioned electronic device.

[0103] The voice processing method provided by the embodiment of the present disclosure, by identifying voice data, obtains the corresponding target recognition text and target voiceprint feature, then determines the target user according to the target voiceprint feature, and then based on the error correction map corresponding to the target user, determines the target information corresponding to the target recognition text. The error correction map includes: the correspondence between the target recognition text and the target information. Finally, the similarity between the target recognition text and the target information is obtained. If the similarity exceeds the preset threshold, the target recognition text is modified to the target information. Through the above method, the voice data of the target user can be corrected, which is beneficial to improving the error correction speed and the user experience.

[0104] To illustrate the present solution in more detail, the following will be described in an exemplary manner in combination with Figure 3A for illustration. It can be understood that Figure 3A the steps involved in the figure may include more steps or fewer steps in actual implementation, and the order of these steps may also be different, subject to the voice processing method provided by the embodiment of the present application.

[0105] Figure 3A It is a schematic flowchart of a voice processing method provided by the embodiment of the present disclosure. Figure 3B It is a schematic diagram of the principle of a voice processing method provided by the embodiment of the present disclosure. This embodiment is applicable to the situation of identifying voice data and modifying the target recognition text. The method of this embodiment can be executed by a voice processing device, and the device can be implemented in a hardware / or software manner and can be configured in an electronic device.

[0106] As Figure 3A shown, the method specifically includes the following steps:

[0107] S310, identify the voice data to obtain the corresponding target recognition text and target voiceprint feature.

[0108] Among them, the target recognition text can be obtained by recognizing speech data through speech recognition technology. The target voiceprint feature can be obtained by recognizing speech data through voiceprint recognition technology. Speech recognition technology is a technology that converts sound signals into text content, usually consisting of two parts: a sound model and a language model. It completes the conversion of sound into factors, and then obtains the optimal text content through the language model. In this disclosure, a speech recognition module is used to convert speech data into the target recognition text. Voiceprint recognition technology is a technology that discriminates user identities through voices. It converts the short-time spectrum of sound signals into Mel-scale Frequency Cepstral Coefficients (MFCC) features, and then uses classification algorithms such as Support Vector Machine (SVM) to output the age group features of the voiceprint. In this disclosure, a voiceprint recognition module is used to convert speech data into the target recognition text. In this embodiment, the age groups of voiceprints are mainly divided into four categories: children, youth, adults, and the elderly. Subsequently, taking the user with a child voiceprint age group as an example, the process of text modification will be described.

[0109] Specifically, after receiving the user's speech data, the speech data can be recognized by the speech recognition module and the voiceprint recognition module respectively to obtain the corresponding target recognition text and target voiceprint feature; it can also be recognized by a module that simultaneously has speech recognition and voiceprint recognition functions to obtain the corresponding target recognition text and target voiceprint feature. This embodiment does not limit the specific recognition method.

[0110] S320. Determine the target user according to the target voiceprint feature.

[0111] Among them, the target user can be a user for whom speech data is prone to errors and has a high error rate, such as a child user (preschool child), an elderly user, etc., and it can also depend on the specific situation. This disclosure does not make a limitation.

[0112] After obtaining the target voiceprint feature, since the target voiceprint feature contains age group features, it is possible to determine whether the user corresponding to the speech data is the target user according to the target voiceprint feature. This step mainly describes the situation where the user corresponding to the speech data is the target user. The situation where the user corresponding to the speech data is a non-target user will be described in subsequent steps.

[0113] S330. Determine the target information corresponding to the target recognition text based on the error correction map corresponding to the target user.

[0114] Among them, the error correction graph can be understood as a pre-established graph corresponding to the target user, which can correspond to the identification information of the recognized text recognized from the voice data of the target user. The identification information can be title information, picture information, control instruction information, etc. Suppose the voice data of the target user contains the title information of a TV program, then the error correction graph can be a multimedia content graph; suppose the voice data of the target user contains control instruction information (such as turning on or off, etc.), then the error correction graph can be a control information graph. Multimedia content can include TV dramas, cartoons, variety shows, etc. The error correction graph includes: the correspondence between the target recognized text and the target information. The target information can be understood as the standard information corresponding to the target recognized text, and the target recognized text is a wrong expression of the target information by the user. For example, suppose the target recognized text is XX TV drama (this text is the wrong name of the TV drama), then the target information is the standard name of this TV drama.

[0115] After determining the target user, an error correction graph corresponding to the target user can be obtained. By querying this error correction graph, the target information corresponding to the target recognized text can be determined, and the target information is the content represented by the central node in the error correction graph.

[0116] S340, obtain the similarity between the target recognized text and the target information. If the similarity exceeds the preset threshold, modify the target recognized text to the target information.

[0117] Among them, the preset threshold can be a preset value, and it can also depend on the specific situation, which is not limited in this disclosure.

[0118] After determining the target information, the similarity between the target recognized text and the target information can be calculated by a similarity calculation method. After obtaining the similarity, compare the similarity with the preset threshold. According to the size relationship between the two, it can be determined whether the similarity exceeds the preset threshold. If the similarity exceeds the preset threshold, it means that the similarity between the target recognized text and the target information is relatively high, and the target recognized text is modified to the target information.

[0119] In some embodiments, after obtaining the similarity between the target recognized text and the target information, if the similarity exceeds the preset threshold and modifying the target recognized text to the target information, it further includes:

[0120] Use the target information as a keyword, so that the corresponding device performs corresponding operations according to the keyword.

[0121] Exemplarily, if the voice data corresponding to the target user is a search intent, the target information is used as a search keyword so that the corresponding device performs a corresponding search operation according to the search keyword; if the voice data corresponding to the target user is a control intent, the target information is used as a control keyword so that the corresponding device performs a corresponding control operation according to the control keyword.

[0122] In this embodiment, through the above method, it is possible to ensure that the subsequent device performs corresponding operations, thereby improving the accuracy of search results or control processes and enhancing user satisfaction.

[0123] Figure 4A It is a schematic flowchart of another voice processing method provided by an embodiment of the present disclosure;

[0124] Figure 4B It is a schematic diagram of the principle of another voice processing method provided by an embodiment of the present disclosure. This embodiment mainly describes the process of determining the similarity between the target recognition text and the target information.

[0125] As Figure 4A shown, the method specifically includes the following steps:

[0126] S410, Recognize the voice data to obtain the corresponding target recognition text and target voiceprint feature.

[0127] S420, Determine the target user according to the target voiceprint feature.

[0128] S430, Based on the error correction map corresponding to the target user, determine the target information corresponding to the target recognition text.

[0129] S440, Based on the error correction map, determine the path information corresponding to the target recognition text.

[0130] Among them, the error correction map includes a central node, each sub-node, and the corresponding edges (i.e., connection relationships).

[0131] After determining the target information corresponding to the target recognition text, by querying the connection relationships between the central node and each sub-node in the error correction map and the connection relationships between different sub-nodes, it is possible to determine each target sub-node that generates the target recognition text. According to the connection relationships between each target sub-node, or the connection relationships between the central node and each target sub-node, the path information corresponding to the target recognition text can be determined.

[0132] S450, According to the path information, determine the similarity between the target recognition text and the target information. If the similarity exceeds a preset threshold, modify the target recognition text to the target information.

[0133] After determining the path information corresponding to the target recognition text, the similarity between the target recognition text and the target information can be calculated based on the weight value of the corresponding edge in the path information. If the similarity exceeds the preset threshold, the target recognition text is modified to the target information.

[0134] In this embodiment, based on the error correction graph, the path information corresponding to the target recognition text is determined, and according to the path information, the similarity between the target recognition text and the target information is determined. Since the error correction graph is relatively intuitive, the above method for determining the similarity is simple and fast, which is beneficial to improving the processing efficiency.

[0135] Figure 5A It is a schematic flowchart of a method for establishing an error correction graph provided by an embodiment of the present disclosure; Figure 5B It is a schematic diagram of the principle of establishing an error correction graph provided by an embodiment of the present disclosure. This embodiment mainly describes the process of establishing an error correction graph.

[0136] Such as Figure 5A shown, the method specifically includes the following steps:

[0137] S510, obtain the first word segments included in the corpus information related to the target information and the second word segments included in the label information corresponding to the target information.

[0138] Among them, the corpus information includes the target information and the text information corresponding to the error expression related to the target information. The label information is extended according to the target information. If the target information is the name of a multimedia content, the label information can be the role object, role name, role type, role attribute, etc. in the multimedia content; if the target information is the control information for a smart device, the label information can be the name of the smart device, the function of the smart device, etc. The present disclosure does not limit this.

[0139] Specifically, by performing word segmentation annotation on all the text contents included in the corpus information, all the first word segments included in the text contents can be obtained, and by performing word segmentation annotation on the label information corresponding to the target information, the second word segments included in the label information can be obtained.

[0140] Exemplarily, assuming the corpus information is "ABCD" and "CBAD", then the first word segments after word segmentation annotation can be: "A", "B", "CD", "CB", and "D"; assuming the label information is the role name (Role Name) and role object (Role Object) corresponding to "ABCD" (the name of a certain cartoon), then the second word segments after word segmentation annotation can be: "a", "b", "c", "d", "e", and "f", etc.

[0141] S520, take the target information as the central node, the first word segmentation, the second word segmentation and the generated information as child nodes, the first association relationship between the central node and different child nodes as the edge between the central node and different child nodes, and the second association relationship between each child node as the edge between each child node, to establish an error correction graph.

[0142] A graph is a semantic network that reveals the relationships between entities. It consists of pieces of knowledge, each of which represents a subject-verb-object triple. The subject and object are represented by nodes, describing the entity name, while the predicate is represented by an edge, describing the entity's action. Generative information can be understood as textual information corresponding to incorrect expressions related to the target information in the corpus. This information can be generated based on the first or second participle, or both, depending on the specific situation. The first association relationship is a splitting relationship, while the second association relationship is a generative relationship.

[0143] It should be noted that each edge has direction information.

[0144] Exemplarily, taking the target information as a program name as an example, the obtained tag information and generation information may be shown in Table 1 below.

[0145] Table 1

[0146]

[0147] After obtaining the first participle and the second participle, the target information is used as the central node, the first participle, the second participle and the generated information are used as child nodes, the splitting relationship between the central node and different child nodes is used as the edge between the central node and different child nodes, and the generation relationship between each child node is used as the edge between each child node, so that an error correction graph can be established.

[0148] In the embodiment of the present disclosure, since the corpus information contains text information corresponding to incorrect expressions related to the target information, the error correction map established by the above method is more comprehensive and has stronger practicality, which is conducive to the subsequent modification of the target recognition text into the target information.

[0149] For example, Figure 5C A schematic diagram of a piece of knowledge in the process of establishing an error correction graph provided by an embodiment of the present disclosure. Node 1 and Node 2 are entity names, and the edge from Node 1 to Node 2 represents the node relationship (rel) between Node 1 and Node 2.

[0150] For example, Figure 5D A schematic diagram of an error correction map provided by an embodiment of the present disclosure is shown as follows: Figure 5DShown: Taking "ABCD" as the target information and children as the target users, through the above process of establishing the error correction graph, with "ABCD" as the central node, the first word segmentation, the second word segmentation and the generated information related to "ABCD" as sub-nodes, and the corresponding edges with direction information, the corresponding error correction graph is established.

[0151] In some embodiments, the method further includes:

[0152] Determining the probability corresponding to the first word segmentation according to the frequency corresponding to the syntactic dependency relationship of the first word segmentation, or the part-of-speech frequency of the target word segmentation corresponding to the core relationship in the syntactic dependency relationship;

[0153] Determining the probability corresponding to the second word segmentation according to the initial weight and the weight factor corresponding to the second word segmentation;

[0154] Based on the probability corresponding to the first word segmentation and the probability corresponding to the second word segmentation, respectively determining the weight values of the corresponding edges in the error correction graph.

[0155] Among them, syntactic dependency relationships may include subject-verb relationship (Subject-Verb, SBV), verb-object relationship (Verb-Bbject, VOB), preposition-object relationship (preposition-object, POV), attribute-head relationship (Attribute, ATT), adverbial-head structure (Adverbial, ADV), complement relationship (Complement, CMP), coordinate relationship (Coordinate, COO) and core relationship (Head, HED), etc. The initial weight and the weight factor can be set customarily or determined according to specific situations, and the present disclosure does not make a limitation. For example, the initial weight can be set to a value such as 0.7 or 0.8. The weight factor can be set according to the priority of the second word segmentation. If the priority of the second word segmentation is high, the corresponding weight factor is set to a larger value, such as 0.9; if the priority of the second word segmentation is low, the corresponding weight factor is set to a smaller value, such as 0.2.

[0156] Specifically, by performing syntactic dependency analysis on all the first word segmentations, the syntactic dependency relationship of each first word segmentation can be obtained.

[0157] Exemplarily, the information table obtained by performing word segmentation annotation on the target information (taking the program name as an example, ABCD and EFGH) and performing syntactic dependency analysis on the word segmentations obtained after the word segmentation annotation can be as shown in Table 2 below.

[0158] Table 2

[0159]

[0160] Among them, text in Table 2 represents the text corresponding to the target information; items represents the number of items; deprel represents the dependent-relation; postag represents the part-of-speech-tagging; id represents the position corresponding to the current word segmentation; word represents the current word segmentation.

[0161] Exemplarily, the information table obtained by performing word segmentation and tagging on the text information corresponding to some incorrect expression ways related to the target information (taking the program name as an example) and performing syntactic dependency analysis on the word segments obtained after word segmentation and tagging can be as shown in Table 3 below.

[0162] Table 3

[0163]

[0164] The frequency corresponding to the syntactic dependency relationship can be determined in the following way: count the occurrence frequency of each syntactic dependency relationship in the syntactic dependency relationships corresponding to all the first word segments respectively, and perform normalization processing to construct a syntactic dependency relationship frequency mapping table, as shown in Table 4 below. Specifically, first determine the number of occurrences of each syntactic dependency relationship in all syntactic dependency analyses respectively, and perform frequency calculation to obtain the occurrence frequency (freq1) corresponding to each syntactic dependency relationship in the syntactic dependency analysis; then divide the occurrence frequency corresponding to each syntactic dependency relationship by the sum of the maximum value (freq 1max ) and the minimum value (freq 1min ) of all occurrence frequencies to obtain the corresponding normalized frequency, and take the normalized frequency as the frequency P dep corresponding to each syntactic dependency relationship, as shown in Formula 1.

[0165]

[0166] Table 4

[0167] Syntactic Dependency Occurrence Frequency Normalized Frequency SBV 0.154 0.440 VOB 0.125 0.357 POV 0.046 0.131 ATT 0.348 0.994 ADV 0.038 0.109 CMP 0.035 0.100

[0168] The part-of-speech frequency of the target word segments corresponding to the core relationship in the syntactic dependency relationship can be determined in the following way. Count the probability of the occurrence of the part-of-speech tags of the target word segments corresponding to the core relationship in the syntactic dependency relationships corresponding to all the first word segments respectively, and perform normalization processing to construct a core word part-of-speech probability table, as shown in Table 5 below. Specifically, count the number of occurrences of the part-of-speech of each target word segment in the target word segments, perform frequency calculation to obtain the occurrence frequency (freq2) corresponding to each part-of-speech, and divide the occurrence frequency corresponding to each part-of-speech by the maximum value (freq 2max ) and the minimum value (freq 2min)Take the ratio of their sum to obtain the corresponding normalized probability, and use it as the corresponding probability P of the core word part of speech core , as shown in Formula 2.

[0169]

[0170] Table 5

[0171] Label Meaning Occurrence Frequency Normalized Probability n Common Noun 0.386 0.990 a Adjective 0.065 0.167 v Common Verb 0.142 0.364 m Quantifier 0.004 0.010 c Conjunction 0.006 0.015

[0172] After constructing the syntactic dependency relationship frequency mapping table (Table 4) and the core word part of speech probability table (Table 5), by querying Table 4 or Table 5, the probability corresponding to the first word segmentation can be obtained. Multiply the initial weight of the second word segmentation by the weight factor to determine the probability corresponding to the second word segmentation. After obtaining the probability corresponding to the first word segmentation and the probability corresponding to the second word segmentation, based on the probability corresponding to the first word segmentation, the weight value of the edge between the central node and different child nodes in the error correction graph can be determined, and based on the probability corresponding to the second word segmentation, the weight values of the edges between each child node in the error correction graph can be determined respectively.

[0173] Exemplarily, taking Figure 5D the error correction graph shown as an example, by determining the weight value corresponding to each edge in the error correction graph through the above method, a path conversion query table can be obtained, as shown in Table 6 below:

[0174] Table 6

[0175] Starting Point Path Type Edge Ending Point Weight Value ABCD Split Path ATT A 0.994 ABCD Split Path ATT B 0.994 ABCD Split Path HED CD 0.99 ABCD ​ ​ a 0.64 ​ ​ ​ b 0.72 ​ ​ ​ c 0.8 ​ ​ ​ d 0.8 ​ ​ ​ e 0.8 ​ ​ ​ f 0.7 A ​ ​ B 0.994 B ​ ​ c 0.994 c ​ ​ ​ 0.99 b ​ ​ ​ 0.994 ​ ​ ​ ​ 0.99 d ​ ​ A 0.994 A ​ ​ f 0.994 f ​ ​ ​ 0.99

[0176] Among them, the split path in Table 6 represents the path from the central node to different child nodes (which can also be called the path corresponding to the first association relationship), and the generated path represents the connection path between each child node (which can also be called the path corresponding to the second association relationship). The end point of the edge with "Role Name" and "Role Object" in the split path is the second word segmentation, and its corresponding weight value is determined by multiplying the initial weight of the second word segmentation by the weight factor.

[0177] Among them, in the split path of Table 6: the second word segmentation with the end point of "a" has an initial weight of 0.8 and a weight factor of 0.8; the second word segmentation with the end point of "b" has an initial weight of 0.8 and a weight factor of 0.9; the second word segmentation with the end point of "c" has an initial weight of 0.8 and a weight factor of 1; the second word segmentation with the end point of "d" has an initial weight of 0.8 and a weight factor of 1; the second word segmentation with the end point of "e" has an initial weight of 0.8 and a weight factor of 1; the second word segmentation with the end point of "f" has an initial weight of 0.7 and a weight factor of 1.

[0178] Exemplarily, ​Schematic diagram of the principle for determining the weight value of the corresponding edge in the error correction graph provided by the embodiments of the present disclosure ​ The corresponding steps have been described in the above embodiments and will not be repeated here.

[0179] In some embodiments, determining the probability corresponding to the first word segmentation includes:

[0180] When it is determined according to the syntactic dependency relationship of the first word segmentation that the first word segmentation is a core word, based on the part-of-speech frequency of the core word, determine the probability corresponding to the core word;

[0181] When it is determined according to the syntactic dependency relationship of the first word segmentation that the first word segmentation is a non-core word, based on the frequency corresponding to the syntactic dependency relationship, determine the probability corresponding to the non-core word.

[0182] Specifically, when it is determined according to the syntactic dependency relationship of the first word segmentation that the first word segmentation is a core word, based on the part-of-speech frequency of the core word, specifically, Table 5 can be queried to determine the probability corresponding to the core word; when it is determined according to the syntactic dependency relationship of the first word segmentation that the first word segmentation is a non-core word, based on the frequency corresponding to the syntactic dependency relationship, specifically, Table 4 can be queried to determine the probability corresponding to the non-core word.

[0183] In this embodiment, since the core word is relatively important, the probability of the corresponding first word segmentation is determined through its part-of-speech frequency. Determining the probability of the first word segmentation through the above two situations respectively is beneficial to improving the accuracy and more conforms to the actual situation.

[0184] Exemplarily, ​ Schematic diagram of the principle for determining the probability corresponding to the first word segmentation provided by the embodiments of the present disclosure ​ The corresponding steps have been described in the above embodiments and will not be repeated here.

[0185] In some embodiments, the path information includes: the first path information and the second path information respectively corresponding to the target recognition text under different path types;

[0186] The determining the similarity between the target recognition text and the target information according to the path information may specifically include:

[0187] Determine the first probability value corresponding to the first path type based on the first path information, and determine the second probability value corresponding to the second path type based on the second path information;

[0188] Determine the similarity between the target recognition text and the target information according to the first probability value and the second probability value.

[0189] Among them, the path types can include split paths and generation paths. A split path can be understood as the path from the central node to different child nodes (which can also be called the path corresponding to the first association relationship), and a generation path can be understood as the connection path between each child node (which can also be called the path corresponding to the second association relationship). When the first path type is a split path, the second path type is a generation path; when the first path type is a generation path, the second path type is a split path. The first path information can include the child nodes included in the first path type. The second path information can include the child nodes included in the second path type.

[0190] Specifically, based on the error correction graph, the first path information and the second path information respectively corresponding to the target recognition text under different path types can be determined. Based on the first path information, by querying the path conversion table (Table 6) corresponding to the error correction graph, the first probability values respectively corresponding to the child nodes (i.e., word segments) included in the first path type can be determined, and based on the second path information, by querying the path conversion table (Table 6) corresponding to the error correction graph, the second probability values respectively corresponding to the child nodes included in the second path type can be determined. After obtaining the first probability value and the second probability value, according to the first probability value and the second probability value, the similarity between the target recognition text and the target information can be determined through the following formula.

[0191]

[0192] Among them, S NT represents the similarity between the target recognition text and the target information. The first part is the first probability value corresponding to the target recognition text under the split path, that is: the weighted sum of the weight values corresponding to the child nodes obtained by splitting the target recognition text; p i represents the weight value corresponding to the i-th word segment, which can be obtained by querying Table 6; q i represents the proportion of the length of the i-th word segment in the total length of the target recognition text; the second part is the second probability value corresponding to the target recognition text under the generation path, that is: the weighted sum of the weight values corresponding to each child node when generating the target recognition text; p j represents the weight value corresponding to the j-th word segment, which can be obtained by querying Table 6; q j represents the proportion of the length of the j-th word segment in the total length of the target recognition text. split represents splitting, and genarate represents generating. i and j are determined according to the split path and the generation path of the target recognition text.

[0193] Exemplarily, taking "dAf" as the target recognition text, the split path is "d - A - f"; the generation path is "d - A - f".

[0194] Exemplarily, ​Schematic diagram of a similarity determination method provided by an embodiment of the present disclosure ​ The corresponding steps have been described in the above embodiments and will not be repeated here.

[0195] Exemplarily, Table 7 is an example of determining similarity by the above method, as shown in Table 7 below:

[0196] Table 7

[0197]

[0198] In this embodiment, the similarity is determined by the above method, and the accuracy is relatively high.

[0199] In some embodiments, the path information includes: the segmentation words represented by each sub-node of the target recognition text under the corresponding path and the correlation probability with the target recognition text, and the weight factor corresponding to the segmentation words represented by each sub-node;

[0200] Determining the similarity between the target recognition text and the target information according to the path information includes:

[0201] Determining the similarity between the target recognition text and the target information according to the correlation probability and the corresponding weight factor.

[0202] Specifically, based on the target graph, it is possible to determine the segmentation words represented by each sub-node of the target recognition text under the generation path (or splitting path) and the correlation probability (i.e., the corresponding weight value) with the target recognition text, and the weight factor corresponding to the segmentation words represented by each sub-node (the weight factor can be set customarily). By performing weighted summation on each correlation probability and the corresponding weight factor, the similarity between the target recognition text and the target information can be determined.

[0203] Exemplarily, ​ Schematic diagram of another similarity determination method provided by an embodiment of the present disclosure ​ The corresponding steps have been described in the above embodiments and will not be repeated here.

[0204] In this embodiment, the similarity is determined by the above method, which is simple and fast.

[0205] In some embodiments, the method further includes: if it is determined according to the target voiceprint feature that the user is a non-target user, or the similarity does not exceed the preset threshold, then the target recognition text is processed to obtain a corresponding processing result;

[0206] Based on the processing result, the processing result is modified by a preset voice correction method to obtain a modified text.

[0207] Among them, the preset speech error correction method can include edit distance error correction and pinyin similarity error correction methods, etc., which are not limited in this embodiment.

[0208] Specifically, if it is determined that the user is a non-target user according to the target voiceprint feature, or the similarity does not exceed the preset threshold, the target recognition text is segmented and labeled to obtain the corresponding processing result. Based on this processing result, the processing result is modified by the preset speech error correction method to obtain the modified text.

[0209] Optionally, if the target information corresponding to the target recognition text cannot be determined, the target recognition text can also be processed to obtain the corresponding processing result, and based on the processing result, the processing result is modified by the preset speech error correction method to obtain the modified text.

[0210] In this embodiment, by modifying through this method, the coverage of the speech processing method can be further improved, ensuring that speech processing can be performed in various situations.

[0211] ​ It is a schematic flowchart of another speech processing method provided by an embodiment of the present disclosure. This embodiment is applicable to explaining the entire speech processing process.

[0212] As ​ shown, the method specifically includes the following steps:

[0213] S7001, recognize the speech data to obtain the corresponding target recognition text and target voiceprint feature.

[0214] S7002, determine whether the user is a target user according to the target voiceprint feature.

[0215] If so, execute S7003; if not, execute S7007-S7008.

[0216] S7003, based on the error correction map corresponding to the target user, determine the target information corresponding to the target recognition text.

[0217] S7004, obtain the similarity between the target recognition text and the target information.

[0218] S7005, determine whether the similarity exceeds the preset threshold.

[0219] If so, execute S7006; if not, execute S7007-S7008.

[0220] S7006, modify the target recognition text to the target information.

[0221] S7007, process the target recognition text to obtain the corresponding processing result.

[0222] S7008, based on the processing result, modifies the processing result through a preset voice error correction method to obtain the modified text.

[0223] ​ FIG. 5 is a schematic structural diagram of a voice processing device provided by an embodiment of the present disclosure; the device is configured in an electronic device and can implement the voice processing method described in any embodiment of the present application. The device specifically includes the following:

[0224] An identification module 801, configured to identify voice data to obtain a corresponding target identification text and a target voiceprint feature;

[0225] A first determination module 802, configured to determine a target user according to the target voiceprint feature;

[0226] A second determination module 803, configured to determine target information corresponding to the target identification text based on an error correction map corresponding to the target user, where the error correction map includes: a correspondence between the target identification text and the target information;

[0227] A modification module 804, configured to obtain a similarity between the target identification text and the target information, and if the similarity exceeds a preset threshold, modify the target identification text to the target information.

[0228] ​ FIG. 6 is a schematic structural diagram of a modification module in the voice processing device according to an embodiment of the present disclosure, as ​ shown, the modification module 804 includes:

[0229] A path information determination unit 8041, configured to determine path information corresponding to the target identification text based on the error correction map;

[0230] A similarity determination unit 8042, configured to determine a similarity between the target identification text and the target information according to the path information;

[0231] A modification unit 8043, configured to, if the similarity exceeds a preset threshold, modify the target identification text to the target information.

[0232] As an optional implementation manner of an embodiment of the present disclosure, the path information includes: first path information and second path information respectively corresponding to the target identification text under different path types;

[0233] The similarity determination unit is configured to:

[0234] Determine a first probability value corresponding to a first path type based on the first path information, and determine a second probability value corresponding to a second path type based on the second path information;

[0235] Determine the similarity between the target recognition text and the target information according to the first probability value and the second probability value.

[0236] As an optional implementation manner of the embodiments of the present disclosure, the path information includes: the relevant probability of the word segmentation represented by each child node of the target recognition text under the corresponding path and the weight factor corresponding to the word segmentation represented by each child node;

[0237] The similarity determination unit is further configured to: determine the similarity between the target recognition text and the target information according to the relevant probability and the corresponding weight factor.

[0238] As an optional implementation manner of the embodiments of the present disclosure, the apparatus further includes: an error correction graph building module, configured to:

[0239] Obtain the first word segmentation included in the corpus information related to the target information and the second word segmentation included in the label information corresponding to the target information;

[0240] Use the target information as the central node, the first word segmentation, the second word segmentation, and the generated information as child nodes, use the first association relationship between the central node and different child nodes as the edge between the central node and different child nodes, and use the second association relationship between each child node as the edge between each child node to build the error correction graph.

[0241] As an optional implementation manner of the embodiments of the present disclosure, the apparatus further includes: a weight value determination module, configured to:

[0242] Determine the probability corresponding to the first word segmentation according to the frequency corresponding to the syntactic dependency relationship of the first word segmentation, or the part-of-speech frequency of the target word segmentation corresponding to the core relationship in the syntactic dependency relationship;

[0243] Determine the probability corresponding to the second word segmentation according to the initial weight and the weight factor corresponding to the second word segmentation;

[0244] Based on the probability corresponding to the first word segmentation and the probability corresponding to the second word segmentation, respectively determine the weight values of the corresponding edges in the error correction graph.

[0245] As an optional implementation manner of the embodiments of the present disclosure, the apparatus further includes:

[0246] A processing module, configured to, if it is determined according to the target voiceprint feature that the user is a non-target user, or the similarity does not exceed the preset threshold, process the target recognition text to obtain a corresponding processing result;

[0247] A text determination module, configured to modify the processing result by a preset speech error correction method based on the processing result, so as to obtain a modified text.

[0248] The speech processing apparatus provided by the embodiments of the present disclosure can execute the speech processing method provided by any embodiment of the present disclosure, and has the corresponding functional modules and beneficial effects for executing the method. To avoid repetition, details are not described herein again.

[0249] An embodiment of the present disclosure provides an electronic device, including: one or more processors; a storage device, configured to store one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors implement any one of the speech processing methods in the embodiments of the present disclosure.

[0250] ​ is a schematic structural diagram of an electronic device provided by an embodiment of the present disclosure. As ​ shown, the electronic device includes a processor 910 and a storage device 920; the number of processors 910 in the electronic device may be one or more, ​ and here one processor 910 is taken as an example; the processor 910 and the storage device 920 in the electronic device may be connected through a bus or other means, ​ and here the connection through the bus is taken as an example.

[0251] The storage device 920, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules, such as program instructions / modules corresponding to the speech processing method in the embodiments of the present disclosure. The processor 910 executes various functional applications and data processing of the electronic device by running the software programs, instructions, and modules stored in the storage device 920, that is, implements the speech processing method provided by the embodiments of the present disclosure.

[0252] The storage device 920 may mainly include a program storage area and a data storage area. Among them, the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created according to the use of the terminal, etc. In addition, the storage device 920 may include a high-speed random access memory, and may further include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state storage devices. In some instances, the storage device 920 may further include a memory remotely disposed relative to the processor 910, and these remote memories may be connected to the electronic device through a network. Examples of the above networks include, but are not limited to, the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0253] The electronic device provided in this embodiment can be used to execute the speech processing method provided in any of the above embodiments, and has corresponding functions and beneficial effects.

[0254] An embodiment of the present disclosure provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements each process executed by the above voice processing method and can achieve the same technical effects. To avoid repetition, it will not be elaborated here.

[0255] Among them, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, an optical disk, or the like.

[0256] For ease of explanation, the above description has been made in conjunction with specific embodiments. However, the above discussion in some embodiments is not intended to be exhaustive or to limit the embodiments to the specific forms disclosed above. According to the above teachings, various modifications and variations can be obtained. The selection and description of the above embodiments are for better explaining the principles and practical applications, so that those skilled in the art can better use the embodiments and various different modified embodiments suitable for specific use considerations.

Claims

1. A voice processing method, characterized in that, The method includes: Performing recognition on the voice data to obtain a corresponding target recognition text and target voiceprint features; Determining a target user according to the target voiceprint features; Based on an error correction graph corresponding to the target user, determining target information corresponding to the target recognition text, where the error correction graph includes: the corresponding relationship between the target recognition text and the target information; Based on the error correction graph, determining path information corresponding to the target recognition text; the path information includes: the word segmentation represented by each sub-node in the corresponding path of the target recognition text, the correlation probability between the word segmentation and the target recognition text, and the weight factor corresponding to the word segmentation represented by each sub-node; According to the correlation probability and the corresponding weight factor, determining the similarity between the target recognition text and the target information, and if the similarity exceeds a preset threshold, modifying the target recognition text to the target information.

2. The method according to claim 1, wherein The path information includes: first path information and second path information respectively corresponding to the target recognition text under different path types; The determining the similarity between the target recognition text and the target information according to the path information includes: Determining a first probability value corresponding to a first path type based on the first path information, and determining a second probability value corresponding to a second path type based on the second path information; According to the first probability value and the second probability value, determining the similarity between the target recognition text and the target information.

3. The method according to claim 1, characterized in that, It further includes: Obtaining a first word segmentation included in corpus information related to the target information and a second word segmentation included in label information corresponding to the target information; Using the target information as a central node, the first word segmentation, the second word segmentation, and generated information as sub-nodes, using a first association relationship between the central node and different sub-nodes as an edge between the central node and different sub-nodes, and using a second association relationship between each sub-node as an edge between each sub-node, to establish the error correction graph.

4. The method according to claim 3, wherein It further includes: Determining the probability corresponding to the first word segmentation according to the frequency corresponding to the syntactic dependency relationship of the first word segmentation, or the part-of-speech frequency of the target word segmentation corresponding to the core relationship in the syntactic dependency relationship; Determining the probability corresponding to the second word segmentation according to the initial weight and weight factor corresponding to the second word segmentation; Based on the probability corresponding to the first word segmentation and the probability corresponding to the second word segmentation, respectively determining the weight value of the corresponding edge in the error correction graph.

5. The method according to any one of claims 1 to 4, characterized in that It further includes: If it is determined according to the target voiceprint features that the user is a non-target user, or the similarity does not exceed the preset threshold, processing the target recognition text to obtain a corresponding processing result; Based on the processing result, modifying the processing result through a preset voice error correction method to obtain a modified text.

6. A voice processing device, characterized in that, The device includes: An identification module, configured to perform recognition on voice data to obtain a corresponding target recognition text and target voiceprint features; A first determination module, configured to determine a target user according to the target voiceprint features; A second determination module, configured to determine target information corresponding to the target recognition text based on an error correction map corresponding to the target user, where the error correction map includes: a correspondence between the target recognition text and the target information; A modification module, configured to determine path information corresponding to the target recognition text based on the error correction map; the path information includes: the segmentation words represented by each child node of the target recognition text under the corresponding path, the correlation probability between the segmentation words and the target recognition text, and the weight factor corresponding to the segmentation words represented by each child node; determine the similarity between the target recognition text and the target information according to the correlation probability and the corresponding weight factor, and if the similarity exceeds a preset threshold, modify the target recognition text to the target information.

7. An electronic device, characterized in that, Comprising: one or more processors; a storage device, configured to store one or more programs, when the one or more programs are executed by the one or more processors, enabling the one or more processors to implement the method according to any one of claims 1-5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by a processor, it implements the method according to any one of claims 1-5.

Images