Hotword management method, speech recognition method, device, apparatus, and storage medium

By acquiring and editing speech recognition text, identifying hot word candidates and updating the hot word database, the problem of inaccurate recognition of professional terms in speech recognition is solved, and personalized, highly accurate speech transcription is achieved.

CN116052676BActive Publication Date: 2026-11-10IFLYTEK CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202211678465.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-12-26
Publication Date
2026-11-10
Estimated Expiration
2042-12-26

AI Technical Summary

Technical Problem

Existing speech recognition technology is not accurate enough when recognizing technical terms. Users need to manually modify the text, but they cannot accumulate data, resulting in poor recognition results in the future.

Method used

By acquiring the first text from speech recognition and the edited second text, candidate hot words are determined, and the hot word library is updated based on usage in the hot word library and user's historical speech data. Target hot words are then selected to improve the accuracy of speech-to-text transcription.

Benefits of technology

By accumulating data and managing personalized hot words, the accuracy of speech-to-text transcription has been improved, conforming to user habits and enhancing the performance of speech recognition.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116052676B_ABST
    Figure CN116052676B_ABST
Patent Text Reader

Abstract

The application provides a hot word management method, a speech recognition method, a device, equipment and a storage medium. The specific implementation scheme is as follows: a first text and a second text of speech recognition are acquired, wherein the second text is obtained by editing the first text; a hot word candidate word is determined in the second text according to an editing type of editing the first text by using the second text; a target hot word is determined in the hot word candidate word by using a hot word library and a use condition of the hot word candidate word in user historical speech data; the hot word library is updated according to the target hot word, and an updated hot word library is obtained; wherein the hot word library is used for assisting speech recognition of to-be-recognized speech. According to the technical scheme of the application, the accuracy of speech transcription text can be effectively and significantly improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of speech recognition technology, and in particular to a hot word management method, speech recognition method, device, equipment and storage medium. Background Technology

[0002] In recent years, with the rapid development of voice technology, speech recognition technology has been widely applied in various industries and fields, including production and daily life, due to its advantages of intelligence, efficiency, and user-friendliness. In the application of speech-to-text technology, the accuracy of transcription is the most basic requirement.

[0003] Currently, there are many speech recognition hardware and software products on the market, which are gradually entering personal work and study as consumer products. However, speech recognition products do not perform well for some technical terms, requiring users to manually edit the transcribed text. But the edited text does not accumulate as data, so it still cannot accurately transcribe speech into text in subsequent speech recognition tasks. Summary of the Invention

[0004] To address the aforementioned issues, this application proposes a hot word management method, a speech recognition method, a device, an equipment, and a storage medium, which can significantly improve the accuracy of speech-to-text transcription.

[0005] According to a first aspect of the embodiments of this application, a hot word management method is provided, including:

[0006] Obtain a first text and a second text from speech recognition, wherein the second text is obtained by editing the first text;

[0007] Based on the type of editing performed on the first text, hot word candidate words are determined in the second text;

[0008] Based on the hot word library and the usage of the hot word candidate words in the user's historical voice data, the target hot word is determined from the hot word candidate words;

[0009] The hot word library is updated based on the target hot words to obtain the updated hot word library; wherein, the hot word library is used to assist in speech recognition of the speech to be recognized.

[0010] According to a second aspect of the embodiments of this application, a speech recognition method is provided, comprising:

[0011] Based on a pre-built hot word library, speech recognition is performed on the speech to be recognized to obtain the speech recognition result;

[0012] The hot word library includes text words selected from the text after the user edits and corrects the historical speech recognition results, which correspond to the user's editing and correction operations.

[0013] According to a third aspect of the embodiments of this application, a hot word management device is provided, comprising:

[0014] The acquisition module is used to acquire the first text and the second text from the speech recognition, wherein the second text is obtained by editing the first text;

[0015] The determining module is used to determine hot word candidate words in the second text based on the type of editing performed on the first text;

[0016] The selection module is used to determine the target hot word from the hot word candidate words based on the hot word library and the usage of the hot word candidate words in the user's historical voice data.

[0017] An update module is used to update the hot word library according to the target hot words to obtain an updated hot word library; wherein, the hot word library is used to assist in speech recognition of the speech to be recognized.

[0018] According to a fourth aspect of the embodiments of this application, a voice recognition device is provided, comprising:

[0019] The recognition module is used to perform speech recognition on the speech to be recognized based on a pre-built hot word library, and obtain the speech recognition result.

[0020] The hot word library includes text words selected from the text after the user edits and corrects the historical speech recognition results, which correspond to the user's editing and correction operations.

[0021] The fifth aspect of this application provides an electronic device, comprising:

[0022] Memory and processor;

[0023] The memory is connected to the processor and is used to store programs;

[0024] The processor implements the aforementioned hot word management method or speech recognition method by running the program in the memory.

[0025] The sixth aspect of this application provides a storage medium storing a computer program, which, when executed by a processor, implements the aforementioned hot word management method or speech recognition method.

[0026] One embodiment of the above application has the following advantages or beneficial effects:

[0027] By utilizing the editing type of the first text, candidate hot words are identified in the second text. Based on the hot word library and the usage of these candidate hot words in the user's historical speech data, target hot words are determined from the candidate hot words. The hot word library is then updated based on these target hot words, resulting in an updated hot word library. In this way, data can be accumulated from the edited text, and personalized target hot words that better match user habits can be selected from the edited text. The hot word library is then updated based on these target hot words. Thus, using the updated hot word library to assist in speech recognition can improve the accuracy of speech-to-text transcription. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0029] Figure 1 A flowchart illustrating a hot word management method provided in an embodiment of this application;

[0030] Figure 2 A flowchart illustrating another hot word management method provided for embodiments of this application;

[0031] Figure 3 A schematic diagram illustrating the specific process of processing edited text for different editing types, provided for embodiments of this application;

[0032] Figure 4 A flowchart illustrating another hot word management method provided in this application embodiment;

[0033] Figure 5 This is a schematic diagram illustrating the specific process of determining target hot words from candidate hot words provided in the embodiments of this application;

[0034] Figure 6 This is a schematic diagram illustrating the specific process of managing hot words in an embodiment of this application;

[0035] Figure 7 A flowchart illustrating a speech recognition method provided in an embodiment of this application;

[0036] Figure 8 This is a schematic diagram of the structure of a hot word management device provided in an embodiment of this application;

[0037] Figure 9 This is a schematic diagram of another speech recognition device provided in an embodiment of this application;

[0038] Figure 10This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0039] The technical solutions of this application are applicable to various speech recognition scenarios, such as meeting scenarios and online education scenarios. Using the technical solutions of this application can improve the accuracy of speech-to-text transcription.

[0040] The technical solutions of this application can be applied, by way of example, to hardware devices such as processors, electronic devices, and servers (including cloud servers), or packaged as software programs and run. When the hardware device executes the processing procedure of the technical solutions of this application, or when the aforementioned software program is run, the purpose of updating the hot word library with personalized hot words that conform to user habits can be achieved. This application only provides an exemplary description of the specific processing procedure of the technical solutions of this application, and does not limit the specific implementation form of the technical solutions of this application. Any technical implementation form that can execute the processing procedure of the technical solutions of this application can be adopted by this application.

[0041] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0042] Exemplary methods

[0043] Figure 1 This is a flowchart of a hot word management method according to an embodiment of this application. In an exemplary embodiment, a hot word management method is provided, including:

[0044] S110. Obtain the first text and the second text for speech recognition, wherein the second text is obtained by editing the first text;

[0045] S120. Based on the editing type of the first text, determine hot word candidate words in the second text;

[0046] S130. Based on the hot word library and the usage of the hot word candidate words in the user's historical voice data, determine the target hot word from the hot word candidate words;

[0047] S140. Update the hot word library according to the target hot words to obtain the updated hot word library; wherein, the hot word library is used to assist in speech recognition of the speech to be recognized.

[0048] In step S110, for example, the first text represents the text obtained from speech recognition. Speech recognition is a method of converting speech into text. Optionally, commonly used speech recognition methods include: linguistic and acoustic methods, stochastic model methods, methods using artificial neural networks, and probabilistic syntax analysis. Specifically, the first text can be obtained by recognizing speech using various open-source acoustic models and language models (such as Markov models). Optionally, it can be text transcribed from speech output by a user in any scenario (e.g., a meeting scenario). The first text and the second text can be words, phrases, sentences, etc., where words can consist of at least one character. The second text is obtained by the user editing the first text according to the actual situation, that is, the second text is text that conforms to the user's habits.

[0049] Optionally, the second text can be obtained by modifying the first text; specifically, it can be by modifying one or more words. Optionally, one or more words can be added to the first text. Optionally, the second text can be obtained by deleting content from the first text.

[0050] In this embodiment, within a meeting scenario, the user's voice is acquired, recognized, and used to obtain first text. This first text is then sent to the user, who edits it. The system can determine whether errors have been corrected by recognizing mouse focus movement or the time interval between mouse focus stays at a fixed position, thus obtaining the first and second texts before and after editing. Alternatively, the system can obtain the first and second texts before and after editing after receiving an editing completion instruction from the user (e.g., the user clicking the "complete" option on the display interface).

[0051] In step S120, for example, hot word candidate words are used to represent potential popular words for the current user, and popular words are used to represent words frequently used by the current user. Editing types include: adding text, deleting text, and modifying text. Optionally, if the number of words in the first text is less than the number of words in the second text, the editing type is deleting text. If the number of words in the first text is greater than the number of words in the second text and the number of newly added words is within a preset quantity threshold, the editing type is adding text, where the preset quantity threshold is determined based on the number of words in the first text. Text editing situations other than adding text and deleting text are determined as modifying text.

[0052] For example, after segmenting the first and second texts, and comparing the segmented texts, if the edit type of the first text is neither "add text" nor "delete text," then the edit type is considered "modify text." Then, candidate hot words are determined based on the modified content. Here, "modified content" refers to text content in the second text that differs from the first text.

[0053] In step S130, for example, the hot word library represents a database composed of multiple popular words for the current user. Optionally, there can be one hot word library per user, or a user can set up multiple hot word libraries, where the multiple hot word libraries are set according to different scenarios. Optionally, it can be text converted from the current user's voice data collected in any scenario, and the current user edits the converted text and selects words to form the hot word library based on the edited content. User historical voice data represents the voice data of the current user obtained in any scenario before the current moment.

[0054] The usage of hot word candidate words in the user's historical voice data can be the frequency of occurrence of hot word candidate words in the first text. Optionally, the frequency of occurrence in the user's historical voice data can be the number of times hot word candidate words appear in the first text, or the number of times the user modifies the text of the speech recognition into hot word candidate words.

[0055] Optionally, hot word candidates can be determined by combining a hot word library and the user's historical voice data. For example, if hot word candidates exist in the hot word library and their frequency in the user's historical voice data is greater than a frequency threshold, then the hot word candidate is identified as the target hot word. This makes the selected target hot word more in line with the user's habits, thereby improving the accuracy of speech recognition.

[0056] Optionally, it can also be the temporal distribution of the candidate hot words, specifically the frequency of use within a specific time period. This specific time period could be the past week, 10:00 AM to 12:00 PM each day, or 1:00 PM to 6:00 PM each day. For example, if a user frequently used word A in their voice recordings over the past week, even if word A was used less frequently than other words in the past month, word A can still be considered a target hot word.

[0057] In step S140, for example, the target hot word may be added to the hot word library to obtain an updated hot word library. Alternatively, the weight of the target hot word may be increased in the hot word library to improve the hit rate of the target hot word.

[0058] In the technical solution of the present application, the editing type for editing the first text is used to determine hot word candidates in the second text; based on the usage of the hot word lexicon and the hot word candidates in the user's historical speech data, target hot words are determined from the hot word candidates, and then the hot word lexicon is updated according to the target hot words to obtain an updated hot word lexicon. In this way, data accumulation can be performed on the edited text, personalized target hot words that better conform to user habits can be selected from the edited text, and the hot word lexicon is updated according to the target hot words, such that speech recognition is assisted based on the updated hot word lexicon, and the accuracy of speech-to-text transcription can be improved.

[0059] In an implementation, the step S120 of determining hot word candidates in the second text according to the editing type for editing the first text includes:

[0060] S210. Comparing the first text and the second text to determine edited text;

[0061] S220. Selecting hot word candidates in the second text according to the editing type for editing the first text and the quantity of the edited text.

[0062] Exemplarily, the edited text is used to represent content of the second text that differs from the first text, and the edited content may be content obtained by modifying the first text, or may also be content newly added to the first text. Optionally, the edited text may be a word, a phrase, or an entire sentence. Optionally, when the content of the second text that differs from the first text is a plurality of words or phrases, each word or phrase may be regarded as one edited text, or a plurality of words or phrases may be regarded as one edited text. Optionally, when multiple parts of the first text are edited, the quantity of edited texts may be determined according to the edited positions in the first text. For example, the first text is "good study, upward", and the second text is "good good study, day day up", where edited text 1 is "good" and edited text 2 is "day day", and the quantity of edited texts at this time is 2.

[0063] Specifically, the first text and the second text are compared to determine the editing type for editing the first text. At least one edited text is determined according to the different content between the two. Corresponding conditions for the editing type and the quantity of edited texts may be pre-stored, and edited texts that satisfy the corresponding conditions are selected from the second text as hot word candidates. In this way, personalized hot word candidates for the user can be determined in the second text according to the editing type and the quantity of the user's edited texts.

[0064] Preferably, in a case where the editing type is a new text addition type, correspondingly, the selecting hot word candidates in the second text according to the editing type for editing the first text and the quantity of the edited text includes:

[0065] If the number of edited texts is not greater than a preset first threshold, the edited texts are identified as hot word candidates.

[0066] For example, such as Figure 3 As shown, when the editing type is "new text," the edited text is considered new text. It is determined whether the number of new text entries exceeds a preset first threshold. If the number of new text entries does not exceed the preset first threshold, it indicates a missed recognition of speech. The new text entries are then extracted, segmented into words, and used as candidate hot words. The preset first threshold can be set according to actual conditions and is not limited here. In this embodiment, the first threshold can be set to 2 new text entries.

[0067] If the number of newly added texts exceeds a preset first threshold, such edits are considered text refinement and optimization, not a missed speech recognition. The newly added texts are discarded without further processing. In this way, by categorizing the editing methods of the first text into different types, missed speech recognitions can be more accurately identified, thereby determining hot word candidate words for the current user.

[0068] Preferably, when the edit type is a text modification type, the step of selecting hot word candidate words in the second text according to the edit type of the first text and the number of edited texts includes:

[0069] If the number of edited texts is not greater than a preset second threshold, the edited texts are identified as hot word candidates.

[0070] If the number of edited texts exceeds a preset second threshold, and if there is a similar text to the edited text in the first text, the edited text is identified as a hot word candidate.

[0071] For example, the preset second threshold is determined based on a preset weighting coefficient and the total number of words in the first text. The preset weighting coefficient can be set according to the actual situation; for example, if the preset weighting coefficient is 50%, then the preset second threshold is the total number of words in the first text * 50%. Approximate text is used to represent words or phrases with similar pronunciations; specifically, matching can be performed based on fuzzy sounds.

[0072] Specifically, such as Figure 3 As shown, when the editing type is "modify text," the edited text is considered modified text. It is then determined whether the number of words in the modified text is greater than 50% of the total number of words in the first text. If the number of words in the modified text is not greater than 50% of the total number of words in the first text, it indicates that the modified content is relatively small, and the words in the modified text can be directly used as candidate hot words.

[0073] If the number of words in the modified text exceeds 50% of the total number of words in the first text, it indicates a significant amount of content has been modified. In this case, the modified text needs to be aligned with the first text to further determine the corresponding content in the first text. Based on the fuzzy phonetic matching rules, words with similar pronunciations to those in the first text (i.e., approximate words) are selected from the modified text and used as hot word candidates. Furthermore, if the edit type is "delete text," the edited text is discarded without further processing.

[0074] In this way, by dividing the editing method of the first text into different editing types, it is possible to more accurately determine the omission of speech recognition or speech recognition error, thereby determining the hot word candidate words for the current user.

[0075] In one implementation, such as Figure 4 As shown, the step S130, which determines the target hot word from the hot word candidate words based on the hot word library and the usage of the hot word candidate words in the user's historical voice data, includes:

[0076] S410. If the hot word candidate exists in the hot word library, the candidate is determined as the target hot word;

[0077] S420. If the hot word candidate does not exist in the hot word library, determine the usage of the hot word candidate in the user's historical voice data, and determine the hot word candidate whose usage meets the set conditions as the target hot word.

[0078] Optionally, step S420 includes: determining the usage time distribution of the hot word candidate words in the user's historical voice data, and determining the hot word candidate words whose usage time distribution meets a preset time threshold as target hot words. The preset time threshold can be any value set according to actual conditions, and is not limited here.

[0079] Specifically, the usage time distribution of hot word candidate words in the past week is obtained. If the usage time of a hot word candidate word exceeds a preset time threshold during the working hours of the past week, the hot word candidate word will be identified as a target hot word even if the frequency of the hot word candidate word is low.

[0080] Preferably, step S420 includes: determining the number of times the hot word candidate words are used in the user's historical voice data, and determining the hot word candidate words whose number of uses is greater than a preset first time threshold as target hot words.

[0081] For example, the number of times a hot word candidate word is used in the user's historical voice data can be the number of times hot word candidate words are searched when performing speech recognition on the historical voice data, or the number of times the historical voice data is transcribed into text and then modified into hot word candidate words. Optionally, the first-time count threshold can be set separately based on the above two different usage counts. The first-time count threshold can be set according to the actual situation and is not limited here.

[0082] Specifically, such as Figure 5 As shown, it is determined whether there are hot word candidates in the hot word library. If there are hot word candidates, it means that the hot word candidate belongs to the current user's hot words, so it is still determined as the target hot word.

[0083] If no candidate hot words exist in the hot word database, but the number of times a candidate hot word is searched during speech recognition of historical speech data is greater than 10, or the number of times the transcribed historical speech data is modified into candidate hot words after being converted to text is greater than 2, then the candidate hot words that meet the above conditions are identified as target hot words. In this way, combining the hot word database with the user's historical speech data allows for more accurate identification of target hot words that align with the current user's habits from the candidate hot words, increasing the accuracy of speech-to-text transcription.

[0084] In one embodiment, the method further includes:

[0085] If the number of times the hot word candidate is used in the user's historical voice data is not greater than a preset first-time threshold and the hot word candidate is a defined type of word, then if there are words in the first text that match the hot word candidate, the hot word candidate is determined as the target hot word; wherein, the words that match the hot word candidate include words that match the pronunciation and / or semantics of the hot word candidate.

[0086] For example, the categorization term is used to represent a word with a specific name. The categorization term can include: technical terms, specific personal names, place names, etc. Optionally, technical terms are used to represent specialized terms in various fields. For example, technical terms in the medical field can include: buffer salts, capillaries, bronchodilators, etc. Technical terms in the biological field can include: macrophages, triglycerides, etc. Optionally, the categorization term can be determined by searching on a website to determine whether hot word candidate words are technical terms or specific personal names or place names. Alternatively, a technical term database can be pre-built, and hot word candidate words can be searched for in the database. In this embodiment, the categorization term uses technical terms.

[0087] Optionally, if the number of times a candidate hot word is used in the user's historical voice data does not exceed a preset first-time threshold and the candidate hot word is a defined type of word, then if there is a word in the first text that matches the pronunciation or semantics of the candidate hot word, the candidate hot word is determined as the target hot word. Specifically, this can be achieved using fuzzy sound matching, or other sound matching methods, which are not limited here. Semantic matching can be performed using neural networks, such as DeepStructured Semantic Models (DSSM). Other methods that can perform semantic analysis on words can also be used, which are not limited here.

[0088] Specifically, such as Figure 5 As shown, when the hot word candidate does not exist in the hot word library, if the number of times the hot word candidate is used in the user's historical voice data does not exceed a preset first-time threshold, that is, the number of times the hot word candidate is searched during speech recognition of historical voice data does not exceed 10 and the number of times the historical voice data is modified into hot word candidate after being transcribed into text does not exceed 2, it is determined whether the hot word candidate is a professional term. If so, the hot word candidate is matched with the first text using fuzzy sound matching. If there are words in the first text that can be matched with fuzzy sound, the hot word candidate that can be matched is determined as the target hot word. If it is not a professional term, it is discarded without further processing. In this way, by combining professional terms with the selection of candidate words, the omission of target hot words is further avoided, and the accuracy of speech-to-text transcription is increased.

[0089] Preferably, the method further includes:

[0090] If the number of times the hot word candidate is used in the user's historical voice data is not greater than a preset first-time threshold and the hot word candidate is a defined type of word, and if there is no word in the first text that matches the hot word candidate, and the hot word candidate exists in the hot word candidate library, then the hot word candidate is determined as the target hot word.

[0091] For example, the hot word candidate library is used to represent a database consisting of hot word candidate words that have not been added to the hot word library. Specifically, such as Figure 5As shown, if the number of times a candidate hot word is used in the user's historical speech data does not exceed a preset threshold (i.e., the number of times a candidate hot word is searched during speech recognition of historical speech data does not exceed 10, and the number of times the historical speech data is modified into a candidate hot word after being transcribed into text does not exceed 2), the system determines whether the candidate hot word is a technical term. If so, it performs fuzzy sound matching between the candidate hot word and the first text. If no fuzzy sound matching words exist in the first text, it checks whether the candidate hot word exists in the candidate hot word library. If it does, the candidate hot word is identified as the target hot word. If it does not exist, the candidate hot word is added to the candidate hot word library. In this way, by combining technical terms and the candidate hot word library to select candidate words, the system further avoids missing target hot words and increases the accuracy of speech-to-text transcription.

[0092] In one implementation, updating the hot word library based on the target hot word to obtain an updated hot word library includes:

[0093] If the target hot word exists in the hot word library, the hot word incentive coefficient of the target hot word is increased in the hot word library.

[0094] For example, such as Figure 5 As shown, the hot word activation coefficient represents the probability of hitting a hot word in the hot word library. Optionally, the hot word activation coefficient can be increased by the same value each time, or it can be a stepped increase. For example, the hot word activation coefficient can be set to 1 each time. Alternatively, when a hot word in the hot word library is selected as a candidate hot word for the first time, the hot word activation coefficient in the hot word library can be increased by 1. If the hot word still cannot be correctly recognized in the next speech recognition, the hot word activation coefficient can be increased by 2.

[0095] Specifically, if a target hot word exists in the hot word library, it means that this target hot word belongs to the current user's hot words. However, the default weight of this hot word in the hot word library is not enough to guarantee the hit effect. Therefore, the hot word incentive coefficient of the target hot word is increased, thereby improving the hit efficiency of hot words.

[0096] In one implementation, after updating the hot word library according to the target hot word to obtain the updated hot word library, the method further includes:

[0097] Speech recognition is performed on the preset test audio set to obtain the first test text set;

[0098] The updated hot word library is used to perform speech recognition on the test audio set to obtain a second test text set;

[0099] If the recognition accuracy of the second test text set is higher than that of the first test text set, speech recognition is performed based on the updated hot word library.

[0100] For example, the preset test audio set includes multiple audio samples pre-collected in any scenario. The test audio set is used to test hot words in the hot word database. Specifically, it can be audio data extracted from web pages, or audio data collected by various sound acquisition devices, without limitation.

[0101] Optionally, a test is performed for each new target hot word added to the hot word library. Before the target hot word is officially loaded and activated, the test audio set can be transcribed using two modes: without loading hot words and with the updated hot word library, resulting in a first test text set and a second test text set, respectively. The accuracy (i.e., general effect) of the recognition in the first and second test text sets is judged. If the accuracy of the second test text set is higher than that of the first test text set, the updated hot word library is officially activated for speech recognition. If the accuracy of the second test text set is not higher than that of the first test text set, the target hot word is removed from the hot word library. In this way, false triggering is verified before the hot word takes effect, ensuring that false triggering does not occur, and data accumulation is formed to improve the subsequent speech recognition effect, making users sticky with the product, and the speech recognition effect becomes more and more accurate with use.

[0102] Optionally, before the target hot words are officially loaded and activated, the test audio set is transcribed using two modes: one without loading hot words and the other with the target hot words from the updated hot word library, resulting in a first test text set and a second test text set, respectively. The accuracy (i.e., general effect) of the recognition in the first and second test text sets is judged. If the accuracy of the second test text set is not higher than that of the first test text set, the target hot word is removed from the hot word library. If the accuracy of the second test text set is higher than that of the first test text set, the updated hot word library is officially activated for speech recognition. Furthermore, in real-time transcription scenarios, the updated hot word library is added to the speech recognition engine to ensure the transcription accuracy of the hot word in subsequent meetings.

[0103] In one embodiment, the method further includes:

[0104] The hot words in the hot word library are updated by using the number of times each hot word is hit and the hit time during the speech recognition process.

[0105] For example, the hit count is used to represent the number of times the hot word is selected during the speech recognition process. The hit time is used to represent the time when the hot word was last selected during the speech recognition process.

[0106] Specifically, the hot word database can be managed based on both the number of hits and the hit time. Hit conditions can be set based on these factors, for example, a minimum of M hits and a maximum hit time of N days, where M and N are positive integers. Hot words that do not meet the hit conditions are removed from the database, while those that do meet the conditions are retained, thus enabling timely management and elimination of hot words.

[0107] Preferably, updating the hot words in the hot word database using the hit count and hit time of each hot word further includes:

[0108] If the hit time of the i-th hot word in the hot word database is within a preset time interval before the current time, the i-th hot word is stored in the hot word database.

[0109] If the hit time of the i-th hot word in the hot word library is not within the preset time interval, and the hit count of the i-th hot word is greater than the second threshold, then the i-th hot word is deleted from the hot word library and added to the hot word candidate library; where i is a positive integer. The preset time interval can be set according to actual needs, such as 1 month, 10 days, etc. The second threshold can be set according to actual needs, such as 5 times, 10 times.

[0110] In this embodiment, as Figure 6 As shown, after each speech recognition operation, the number of hits and the time of hits for hot words in the hot word database are statistically analyzed. If the most recent hit time is greater than one month (a preset time interval) and the number of hits is less than 10 (the second count threshold), the hot word is removed from the hot word database. If the most recent hit time is greater than one month and the number of hits is no less than 10, the hot word is removed from the hot word database and added to the hot word candidate database. If the most recent hit time is no more than one month, the number of hits and the time of hits for the hot word are refreshed.

[0111] Figure 7 This is a flowchart of a speech recognition method according to an embodiment of this application. In an exemplary embodiment, a speech recognition method is provided, including:

[0112] S710: Based on a pre-built hot word library, perform speech recognition on the speech to be recognized and obtain the speech recognition result;

[0113] The hot word library includes text words selected from the text after the user edits and corrects the historical speech recognition results, corresponding to the user's editing and correction operations.

[0114] In one implementation, the hot word library is constructed through the following process:

[0115] Obtain a first text and a second text from speech recognition, wherein the second text is obtained by editing the first text;

[0116] Based on the type of editing performed on the first text, hot word candidate words are determined in the second text;

[0117] Based on the hot word library and the usage of the hot word candidate words in the user's historical voice data, the target hot word is determined from the hot word candidate words;

[0118] The hot word library is updated based on the target hot words to obtain the updated hot word library.

[0119] For example, the hot word library in the above speech recognition method can be obtained from any embodiment of the above hot word management method.

[0120] In the technical solution of this application, speech recognition is performed on the speech to be recognized according to a pre-built hot word library to obtain the speech recognition result. Since the pre-built hot word library includes text words selected from the text after the user edits and corrects the historical speech recognition results, which correspond to the user's editing and correction operations, the hot words in the hot word library conform to the user's habits, making the speech recognition result more accurate.

[0121] Exemplary device

[0122] Correspondingly, Figure 8 This is a schematic diagram of a hot word management device according to an embodiment of this application. In an exemplary embodiment, a hot word management device is provided, including:

[0123] The acquisition module 810 is used to acquire the first text and the second text of the speech recognition, wherein the second text is obtained by editing the first text;

[0124] The determining module 820 is used to determine hot word candidate words in the second text based on the editing type of the first text;

[0125] Selection module 830 is used to determine target hot words from the hot word candidates based on the hot word library and the usage of the hot word candidate words in the user's historical voice data.

[0126] The update module 840 is used to update the hot word library according to the target hot words to obtain the updated hot word library; wherein, the hot word library is used to assist in the speech recognition of the speech to be recognized.

[0127] In one implementation, the determining module includes:

[0128] The comparison module is used to compare the first text and the second text to determine the text to be edited;

[0129] The first processing module is used to select hot word candidate words in the second text based on the editing type of the first text and the number of edited texts.

[0130] In one implementation, when the editing type is new text, the first processing module is further configured to:

[0131] If the number of edited texts is not greater than a preset first threshold, the edited texts are identified as hot word candidates.

[0132] In one implementation, when the editing type is a text modification type, the first processing module is further configured to:

[0133] If the number of edited texts is not greater than a preset second threshold, the edited texts are identified as hot word candidates.

[0134] If the number of edited texts exceeds a preset second threshold, and if there is a similar text to the edited text in the first text, the edited text is identified as a hot word candidate.

[0135] In one implementation, the selection module includes:

[0136] The second processing module is used to determine the candidate hot words as target hot words when the candidate hot words exist in the hot word library;

[0137] The third processing module is used to determine the usage of the hot word candidate in the user's historical voice data when the hot word candidate does not exist in the hot word library, and to determine the hot word candidate whose usage meets the set conditions as the target hot word.

[0138] In one implementation, the third processing module is further configured to: determine the number of times the hot word candidate words are used in the user's historical voice data, and determine the hot word candidate words whose number of uses is greater than a preset first number threshold as target hot words.

[0139] In one embodiment, the device further includes:

[0140] The fourth processing module is used to determine the hot word candidate as a target hot word if there are words in the first text that match the hot word candidate in the user's historical voice data, provided that the number of times the hot word candidate is used in the user's historical voice data is not greater than a preset first-time threshold and the hot word candidate is a word of a set type. The words that match the hot word candidate include words that match the pronunciation and / or semantics of the hot word candidate.

[0141] In one embodiment, the device further includes:

[0142] The fifth processing module is used to determine the hot word candidate as the target hot word when the number of times the hot word candidate is used in the user's historical voice data is not greater than a preset first-time threshold and the hot word candidate is a set type word. If there is no word matching the hot word candidate in the first text and the hot word candidate exists in the hot word candidate library, then the hot word candidate is determined as the target hot word.

[0143] In one implementation, the update module is further configured to:

[0144] If the target hot word exists in the hot word library, the hot word incentive coefficient of the target hot word is increased in the hot word library.

[0145] In one embodiment, the device further includes:

[0146] The first test module is used to perform speech recognition on a preset set of test audio to obtain a first set of test text.

[0147] The second testing module is used to perform speech recognition on the test audio set using the updated hot word library to obtain the second test text set.

[0148] The verification module is used to perform speech recognition based on the updated hot word library when the recognition accuracy of the second test text set is higher than that of the first test text set.

[0149] In one embodiment, the device further includes:

[0150] The management module is used to update the hot words in the hot word library by using the number of times each hot word is hit and the hit time during the speech recognition process.

[0151] In one implementation, the management module is further configured to:

[0152] If the hit time of the i-th hot word in the hot word database is within a preset time interval before the current time, the i-th hot word is stored in the hot word database.

[0153] If the hit time of the i-th hot word in the hot word library is not within the preset time interval, and the hit count of the i-th hot word is greater than the second threshold, then the i-th hot word is deleted from the hot word library and added to the hot word candidate library; where i is a positive integer.

[0154] Correspondingly, Figure 9 This is a schematic diagram of a speech recognition device according to an embodiment of this application. In an exemplary embodiment, a speech recognition device is provided, including:

[0155] The recognition module 910 is used to perform speech recognition on the speech to be recognized based on a pre-built hot word library, and obtain the speech recognition result.

[0156] The hot word library includes text words selected from the text after the user edits and corrects the historical speech recognition results, corresponding to the user's editing and correction operations.

[0157] In one implementation, the hot word library is constructed through the following process:

[0158] Obtain a first text and a second text from speech recognition, wherein the second text is obtained by editing the first text;

[0159] Based on the type of editing performed on the first text, hot word candidate words are determined in the second text;

[0160] Based on the hot word library and the usage of the hot word candidate words in the user's historical voice data, the target hot word is determined from the hot word candidate words;

[0161] The hot word library is updated based on the target hot words to obtain the updated hot word library.

[0162] The apparatus provided in this embodiment belongs to the same concept as the method provided in the above embodiments of this application, and can execute the method provided in any of the above embodiments of this application, possessing the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the specific processing content of the method provided in the above embodiments of this application, and will not be repeated here.

[0163] Exemplary electronic devices

[0164] Another embodiment of this application also provides an electronic device, see [link to relevant documentation] Figure 10 As shown, the device includes:

[0165] Memory 1000 and processor 1010;

[0166] The memory 1000 is connected to the processor 1010 and is used to store programs;

[0167] The processor 1010 is used to implement the hot word management method or speech recognition method disclosed in any of the above embodiments by running the program stored in the memory 1000.

[0168] Specifically, the aforementioned electronic device may also include: a bus, a communication interface 1020, an input device 1030, and an output device 1040.

[0169] The processor 1010, memory 1000, communication interface 1020, input device 1030, and output device 1040 are interconnected via a bus. Among them:

[0170] A bus can include a pathway for transmitting information between various components of a computer system.

[0171] The processor 1010 can be a general-purpose processor, such as a general-purpose central processing unit (CPU), a microprocessor, etc., or an application-specific integrated circuit (ASIC), or one or more integrated circuits used to control the execution of the program of the present invention. It can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0172] The processor 1010 may include a main processor, as well as a baseband chip, modem, etc.

[0173] The memory 1000 stores a program that executes the technical solution of this invention, and may also store an operating system and other key business functions. Specifically, the program may include program code, which includes computer operation instructions. More specifically, the memory 1000 may include read-only memory (ROM), other types of static storage devices capable of storing static information and instructions, random access memory (RAM), other types of dynamic storage devices capable of storing information and instructions, disk storage, flash memory, etc.

[0174] Input device 1030 may include a device for receiving user input data and information, such as a keyboard, mouse, camera, scanner, light pen, voice input device, touch screen, pedometer, or gravity sensor.

[0175] Output device 1040 may include devices that allow information to be output to a user, such as a display screen, printer, speaker, etc.

[0176] The communication interface 1020 may include a device that uses any transceiver to communicate with other devices or communication networks, such as Ethernet, Radio Access Network (RAN), Wireless Local Area Network (WLAN), etc.

[0177] The processor 1010 executes the program stored in the memory 1000 and calls other devices, which can be used to implement any of the steps of the hot word management method or speech recognition method provided in the above embodiments of this application.

[0178] Exemplary computer program products and storage media

[0179] In addition to the methods and devices described above, embodiments of this application may also be computer program products, which include computer program instructions that, when executed by a processor, cause the processor to perform the steps of the hot word management method or speech recognition method according to various embodiments of this application as described in the "Exemplary Methods" section above.

[0180] The computer program product can be written in any combination of one or more programming languages ​​to perform the operations of the embodiments of this application. The programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0181] Furthermore, embodiments of this application may also be storage media storing a computer program, which is executed by a processor in the steps of the hot word management method or speech recognition method according to various embodiments of this application described in the "Exemplary Methods" section above.

[0182] The specific working content of the aforementioned electronic device, as well as the specific working content of the aforementioned computer program product and the computer program on the storage medium being run by the processor, can all be found in the content of the aforementioned method embodiments, and will not be repeated here.

[0183] For the foregoing method embodiments, in order to simplify the description, they are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, because according to this application, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0184] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For apparatus embodiments, since they are basically similar to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0185] The steps in the methods of the various embodiments of this application can be adjusted, merged, or deleted in order according to actual needs, and the technical features described in each embodiment can be replaced or combined.

[0186] The modules and sub-modules in the various embodiments of the present application's devices and terminals can be merged, divided, and deleted according to actual needs.

[0187] It should be understood that the disclosed terminals, devices, and methods can be implemented in other ways, given the several embodiments provided in this application. For example, the terminal embodiments described above are merely illustrative. For instance, the division of modules or sub-modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple sub-modules or modules may be combined or integrated into another module, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or modules, and may be electrical, mechanical, or other forms.

[0188] The modules or submodules described as separate components may or may not be physically separate. The components that constitute a module or submodule may or may not be physical modules or submodules; that is, they may be located in one place or distributed across multiple network modules or submodules. Some or all of the modules or submodules can be selected to achieve the purpose of this embodiment's solution, depending on actual needs.

[0189] Furthermore, the functional modules or sub-modules in the various embodiments of this application can be integrated into one processing module, or each module or sub-module can exist physically separately, or two or more modules or sub-modules can be integrated into one module. The integrated modules or sub-modules described above can be implemented in hardware or in the form of software functional modules or sub-modules.

[0190] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0191] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software unit executed by a processor, or a combination of both. The software unit can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0192] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0193] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for managing trending words, characterized in that, include: Obtain a first text and a second text from speech recognition, wherein the second text is obtained by editing the first text; Based on the editing type of the first text, hot word candidate words are determined in the second text; the editing types include: adding text type, deleting text type, and modifying text type; Based on the hot word library and the usage of the hot word candidate words in the user's historical voice data, target hot words are determined from the hot word candidate words; the usage of the hot word candidate words in the user's historical voice data refers to the frequency of occurrence of the hot word candidate words in the first text; The hot word library is updated based on the target hot words to obtain the updated hot word library; wherein, the hot word library is used to assist in the speech recognition of the speech to be recognized.

2. The method according to claim 1, characterized in that, The step of determining hot word candidate words in the second text based on the editing type of the first text includes: Compare the first text and the second text to determine the text to be edited; Based on the type of editing performed on the first text and the number of texts edited, hot word candidate words are selected in the second text.

3. The method according to claim 2, characterized in that, When the edit type is "new text", the step of selecting hot word candidates in the second text based on the edit type of the first text and the number of edited texts includes: If the number of edited texts is not greater than a preset first threshold, the edited texts are identified as hot word candidates.

4. The method according to claim 2, characterized in that, When the edit type is "modify text", the step of selecting hot word candidates in the second text based on the edit type of the first text and the number of edited texts includes: If the number of edited texts is not greater than a preset second threshold, the edited texts are identified as hot word candidates. If the number of edited texts exceeds a preset second threshold, and if there is a similar text to the edited text in the first text, the edited text is identified as a hot word candidate.

5. The method according to claim 1, characterized in that, The step of determining target hot words from the hot word candidate list based on the hot word library and the usage of the hot word candidate words in the user's historical voice data includes: If the candidate hot words exist in the hot word library, the candidate words are identified as target hot words; If the hot word candidate does not exist in the hot word library, the usage of the hot word candidate in the user's historical voice data is determined, and the hot word candidate whose usage meets the set conditions is determined as the target hot word.

6. The method according to claim 5, characterized in that, Determine the usage of the candidate hot words in the user's historical voice data, and identify the candidate hot words whose usage meets the set conditions as target hot words, including: The number of times the hot word candidate words are used in the user's historical voice data is determined, and hot word candidate words whose number of uses is greater than a preset first time threshold are determined as target hot words.

7. The method according to claim 6, characterized in that, The method further includes: If the number of times the hot word candidate is used in the user's historical voice data is not greater than a preset first-time threshold and the hot word candidate is a defined type of word, then if there are words in the first text that match the hot word candidate, the hot word candidate is determined as the target hot word; wherein, the words that match the hot word candidate include words that match the pronunciation and / or semantics of the hot word candidate.

8. The method according to claim 7, characterized in that, The method further includes: If the number of times the hot word candidate is used in the user's historical voice data is not greater than a preset first-time threshold and the hot word candidate is a defined type of word, then if there is no word in the first text that matches the hot word candidate, and the hot word candidate exists in the hot word candidate library, then the hot word candidate is determined as the target hot word.

9. The method according to claim 5, characterized in that, The step of updating the hot word library based on the target hot word to obtain the updated hot word library includes: If the target hot word exists in the hot word library, the hot word incentive coefficient of the target hot word is increased in the hot word library.

10. The method according to claim 1, characterized in that, After updating the hot word library based on the target hot word to obtain the updated hot word library, the method further includes: Speech recognition is performed on the preset test audio set to obtain the first test text set; The updated hot word library is used to perform speech recognition on the test audio set to obtain a second test text set. If the recognition accuracy of the second test text set is higher than that of the first test text set, speech recognition is performed based on the updated hot word library.

11. The method according to any one of claims 1-10, characterized in that, The method further includes: The hot words in the hot word library are updated by using the number of times each hot word is hit and the hit time during the speech recognition process.

12. The method according to claim 11, characterized in that, The step of updating the hot words in the hot word database by using the hit count and hit time of each hot word in the database also includes: If the hit time of the i-th hot word in the hot word library is within a preset time interval before the current time, the i-th hot word will be saved in the hot word library. If the hit time of the i-th hot word in the hot word library is not within the preset time interval, and the hit count of the i-th hot word is greater than the second threshold, then the i-th hot word is deleted from the hot word library and added to the hot word candidate library; where i is a positive integer.

13. A speech recognition method, characterized in that, include: Based on a pre-built hot word library, speech recognition is performed on the speech to be recognized to obtain the speech recognition result; The hot word library is obtained by performing the hot word management method as described in any one of claims 1 to 12.

14. A hot word management device, characterized in that, include: The acquisition module is used to acquire the first text and the second text from the speech recognition, wherein the second text is obtained by editing the first text; The determining module is used to determine hot word candidate words in the second text based on the editing type of the first text; the editing type includes: adding text type, deleting text type, and modifying text type; The selection module is used to determine target hot words from the hot word candidates based on the user's historical voice data, a hot word library, and the usage of the hot word candidates in the user's historical voice data; the usage of the hot word candidates in the user's historical voice data refers to the frequency of occurrence of the hot word candidates in the first text; An update module is used to update the hot word library according to the target hot words to obtain an updated hot word library; wherein, the hot word library is used to assist in speech recognition of the speech to be recognized.

15. A voice recognition device, characterized in that, include: The recognition module is used to perform speech recognition on the speech to be recognized based on a pre-built hot word library, and obtain the speech recognition result. The hot word library is obtained by performing the hot word management method as described in any one of claims 1 to 12.

16. An electronic device, characterized in that, include: Memory and processor; The memory is connected to the processor and is used to store programs; The processor, by running the program in the memory, implements the hot word management method or speech recognition method as described in any one of claims 1 to 13.

17. A storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the hot word management method or speech recognition method as described in any one of claims 1 to 13.

Citation Information

Patent Citations

  • Voice input method, electronic equipment and computer storage medium

    CN114327355A

  • Automatic speech recognition learning using user corrections

    CN1645477A