Speech recognition result screening method and device, electronic equipment and medium
By screening and analyzing the output of the speech recognition model, determining the text to be marked, solving the problem of manual correction of dependence, saving manpower and improving recognition accuracy.
Patent Information
- Application Number
- CN202310620490.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-29
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2043-05-29
AI Technical Summary
In existing speech recognition technology, correcting inaccurate results of recognition relies on manual annotation, resulting in waste of human resources and the inability to effectively cover new vocabulary.
By obtaining multiple candidate recognition texts and their number of recognitions output by the preset speech recognition model in the historical time period, filter out the target recognition text matching the preset corpus, calculate the number of recognition changes parameters, determine the text to be marked, and reduce manual labeling Quantity and accurately locate text that needs to be corrected.
The number of subsequent voice recognition results labels is reduced, human resources are saved, and the text that needs to be corrected is accurately positioned through the number of recognition changes parameters, improving the recognition accuracy.
Smart Images

Figure CN119049470A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of speech recognition technology, and in particular to a method, device, electronic device and medium for screening speech recognition results. Background Art
[0002] In the process of recognizing speech through a speech recognition model, some inaccurate recognition is inevitable. By correcting the inaccurate recognition results in the speech recognition results, the recognition accuracy of subsequent speech recognition models can be improved.
[0003] In the related art, there is a method of manually annotating and correcting the recognition results. However, the method of manually annotating and correcting has the problem of being labor-intensive. Summary of the invention
[0004] In order to overcome the problems existing in the related art, the present disclosure provides a method, device, electronic device and medium for screening speech recognition results.
[0005] According to a first aspect of an embodiment of the present disclosure, a method for screening speech recognition results is provided, the method for screening speech recognition results comprising:
[0006] Acquire multiple candidate recognition texts output by a preset speech recognition model in a first historical time period, and first recognition times corresponding to the multiple candidate recognition texts respectively. The first recognition time corresponding to a candidate recognition text represents the number of times the preset speech recognition model outputs the candidate recognition text in the first historical time period;
[0007] Screening the plurality of candidate recognition texts to obtain a first target recognition text that is identical to a text in a preset corpus, wherein the text in the preset corpus is an accurately recognized text corresponding to the speech data to be recognized;
[0008] For any first target recognition text among the first target recognition texts, based on the first recognition number corresponding to the first target recognition text and the second recognition number corresponding to the first target recognition text, a recognition number change parameter corresponding to the first target recognition text is obtained, wherein the second recognition number corresponding to the first target recognition text represents the number of times the preset speech recognition model outputs the first target recognition text in a time period before the first historical time period;
[0009] Based on the recognition times variation parameters corresponding to the first target recognition texts, the recognition texts to be annotated are determined from the first target recognition texts.
[0010] Optionally, the time period before the first historical time period includes a second historical time period before the first historical time period and adjacent to the first historical time period and having the same time length, and the obtaining of the recognition number change parameter corresponding to the first target recognition text based on the first recognition number corresponding to the first target recognition text and the second recognition number corresponding to the first target recognition text includes:
[0011] Based on the first recognition times corresponding to the first target recognition text and the second recognition times corresponding to the first target recognition text in the second historical time period, a recognition times change parameter corresponding to the first target recognition text is obtained.
[0012] Optionally, obtaining the recognition times change parameter corresponding to the first target recognition text based on the first recognition times corresponding to the first target recognition text and the second recognition times corresponding to the first target recognition text in the second historical time period includes:
[0013] Determine a difference between the first recognition count corresponding to the first target recognition text and the second recognition count corresponding to the first target recognition text in the second historical time period;
[0014] Determine the difference as the recognition times change parameter corresponding to the first target recognition text, and / or determine the ratio of the difference to the second recognition times corresponding to the first target recognition text in the second historical time period as the recognition times change parameter corresponding to the first target recognition text;
[0015] The determining the to-be-annotated recognition text from each of the first target recognition texts based on the recognition number change parameters corresponding to each of the first target recognition texts includes:
[0016] From each of the first target recognition texts, a preset number of first target recognition texts ranked higher in the corresponding recognition number change parameter are obtained as the recognition texts to be annotated, and / or a preset number of first target recognition texts ranked lower in the corresponding recognition number change parameter are obtained as the recognition texts to be annotated.
[0017] Optionally, the first historical time period corresponds to a preset duration, the time period before the first historical time period includes a second historical time period before the first historical time period and adjacent to the first historical time period and having the preset duration, and includes a third historical time period before the first historical time period and adjacent to the first historical time period and having a duration greater than the preset duration, and the obtaining of a recognition times change parameter corresponding to the first target recognition text based on the first recognition times corresponding to the first target recognition text and the second recognition times corresponding to the first target recognition text includes:
[0018] Based on the first recognition times corresponding to the first target recognition text, the second recognition times corresponding to the first target recognition text in the second historical time period, and the third recognition times corresponding to the first target recognition text in the third historical time period, the recognition times change parameter corresponding to the first target recognition text is obtained.
[0019] Optionally, obtaining the recognition number change parameter corresponding to the first target recognition text based on the first recognition number corresponding to the first target recognition text, the second recognition number corresponding to the first target recognition text in the second historical time period, and the third recognition number corresponding to the first target recognition text in the third historical time period includes:
[0020] The product of a first value and a second value is determined as a recognition times change parameter corresponding to the first target recognition text, wherein the first value is a ratio of the first recognition times corresponding to the first target recognition text to the corresponding third recognition times, and the second value is obtained based on the second recognition times corresponding to the first target recognition text and a preset logarithmic function;
[0021] The determining the to-be-annotated recognition text from each of the first target recognition texts based on the recognition number change parameters corresponding to each of the first target recognition texts includes:
[0022] From each of the first target recognition texts, a preset number of first target recognition texts that are ranked top in the corresponding recognition times change parameter are obtained as the recognition texts to be annotated.
[0023] Optionally, the method further comprises:
[0024] According to the first recognition times corresponding to the plurality of candidate recognition texts, respectively, determining a preset number of third target recognition texts whose corresponding first recognition times are ranked top from the plurality of candidate recognition texts;
[0025] According to the third target recognition text, a recognition text to be processed is determined.
[0026] Optionally, the method further comprises:
[0027] The multiple candidate recognition texts are screened to obtain a second target recognition text that is different from the text in the preset corpus, and the second target recognition text is used for manual recognition and annotation.
[0028] Optionally, the method further comprises:
[0029] Acquire the accurately recognized text of the speech data corresponding to the second target recognized text;
[0030] The accurately recognized text is added to the preset corpus.
[0031] According to a second aspect of an embodiment of the present disclosure, a device for filtering speech recognition results is provided, the device for filtering speech recognition results comprising:
[0032] An acquisition module is configured to acquire a plurality of candidate recognition texts output by a preset speech recognition model in a first historical time period, and first recognition times corresponding to the plurality of candidate recognition texts, respectively. The first recognition time corresponding to a candidate recognition text represents the number of times the preset speech recognition model outputs the candidate recognition text in the first historical time period;
[0033] A screening module is configured to screen the plurality of candidate recognition texts to obtain a first target recognition text that is the same as a text in a preset corpus, wherein the text in the preset corpus is an accurately recognized text corresponding to the speech data to be recognized;
[0034] A calculation module is configured to obtain, for any first target recognition text among the first target recognition texts, a recognition number change parameter corresponding to the first target recognition text based on the first recognition number corresponding to the first target recognition text and the second recognition number corresponding to the first target recognition text, wherein the second recognition number corresponding to the first target recognition text represents the number of times the preset speech recognition model outputs the first target recognition text in a time period before the first historical time period;
[0035] The determination module is configured to determine the recognition text to be marked from each of the first target recognition texts based on the recognition number change parameters corresponding to each of the first target recognition texts.
[0036] According to a third aspect of an embodiment of the present disclosure, an electronic device is provided, comprising: a processor; a memory for storing processor executable instructions; wherein the processor is configured to implement the steps of the method described in the first aspect when executed.
[0037] According to a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided, on which computer program instructions are stored. When the program instructions are executed by a processor, the steps of the method provided in the first aspect of the present disclosure are implemented.
[0038] The present disclosure provides a method, device, electronic device and medium for screening speech recognition results. The method obtains multiple candidate recognition texts output by a preset speech recognition model in a first historical time period, and the first recognition times corresponding to the multiple candidate recognition texts, and screens the multiple candidate recognition texts to obtain first target recognition texts that are the same as the texts in the preset corpus. For any first target recognition text among the first target recognition texts, based on the first recognition times corresponding to the first target recognition text and the second recognition times corresponding to the first target recognition text, a recognition times change parameter corresponding to the first target recognition text is obtained. Based on the recognition times change parameters corresponding to the first target recognition texts, the recognition texts to be annotated are determined from the first target recognition texts. The recognition texts to be annotated are obtained by mining the recognition times change parameters corresponding to the first target recognition texts. On the one hand, the number of annotations of subsequent speech recognition results can be reduced, saving human resources. On the other hand, the recognition texts that need to be annotated can be accurately located.
[0039] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.
[0041] Figure 1 is a flow chart of a method for screening speech recognition results according to an exemplary embodiment of the present disclosure;
[0042] Figure 2 is a flow chart of a method for screening speech recognition results according to an exemplary embodiment of the present disclosure;
[0043] Figure 3 is a schematic diagram of visual display of a TOP text according to an exemplary embodiment of the present disclosure;
[0044] Figure 4 is a block diagram of a device for screening speech recognition results according to an exemplary embodiment;
[0045] Figure 5 The diagram is a schematic structural diagram of an electronic device according to an exemplary embodiment. DETAILED DESCRIPTION
[0046] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.
[0047] It should be noted that all actions of acquiring signals, information or data in this application are carried out in compliance with the relevant data protection laws and policies of the country where they are located, and with the authorization given by the owner of the corresponding device.
[0048] The speech recognition model needs to continuously discover recognition errors in its recognition results so that it can be optimized in subsequent iterations to improve the recognition accuracy of subsequent speech recognition models.
[0049] In the related art, there is a method of correcting recognition results through an NLP (Natural Language Processing) model. The NLP model can determine whether the recognition result is correct and correct the recognition result with error. However, since the NLP model can only correct the trained text results, it cannot achieve comprehensive coverage, for example, it cannot cover some newly added words.
[0050] In addition to correcting the recognition results through the NLP model, the recognition results can also be corrected manually. However, since the speech recognition model deployed for application collects large quantities of data, manual labeling of large quantities of data cannot be operated on a regular basis in terms of time and labor costs.
[0051] In order to solve the above problems, an embodiment of the present disclosure provides a method for screening speech recognition results. Figure 1 is a flow chart of a method for screening speech recognition results according to an exemplary embodiment. Figure 1 As shown, the method for screening speech recognition results is applied to an electronic device, such as a mobile phone, a computer, a tablet device, etc. The method for screening speech recognition results may include the following steps:
[0052] Step S110, obtaining a plurality of candidate recognition texts output by a preset speech recognition model within a first historical time period, and first recognition times corresponding to the plurality of candidate recognition texts.
[0053] Among them, the first recognition count corresponding to a candidate recognition text represents the number of times the preset speech recognition model outputs the candidate recognition text within the first historical time period.
[0054] In the disclosed embodiments, the speech recognition model may be a speech recognition model deployed online or offline, for example, the “Xiao Ai” speech recognition model.
[0055] The first historical time period may be a time period such as one day, one week, or one month before the current time.
[0056] It can be understood that in order to improve the recognition accuracy of the preset voice recognition model, the voice recognition model can, with the user's authorization, collect the voice recognition input data and output recognition results of each user each time. Thus, the electronic device in the embodiment of the present disclosure can obtain the input data and recognition results of the preset voice recognition model within the first historical time period, so as to be able to count the number of input data and recognition results.
[0057] Furthermore, the electronic device can count the various recognition results output by the preset speech recognition model within the first historical time period to obtain the number of recognition times corresponding to each recognition result, that is, if a text is recognized once, the number of recognition times is counted once.
[0058] For example, during the speech recognition process, Xiao Ai outputs the recognition result "Play Tashanhe" for a certain input speech, then the corresponding recognition result "Play Tashanhe" accumulates one recognition count.
[0059] In some embodiments, for some speech recognition models with small-scale applications, considering the small amount of data, all recognition results of the speech recognition model within the first historical time period can be obtained. All the obtained recognition results are used as multiple candidate recognition texts, and the recognition number corresponding to each candidate recognition result is used as the first recognition number.
[0060] In other embodiments, for some widely used speech recognition models, such as the "Xiao Ai" speech recognition model, the amount of data generated every day is huge. Therefore, all recognition results of the speech recognition model within the first historical time period can be obtained, and each recognition result can be sorted according to the number of recognition times of each recognition result in the first historical time period, so as to obtain a preset number of recognition results with the top corresponding recognition times as the multiple candidate recognition texts obtained, and the recognition times corresponding to each candidate recognition result are taken as the first recognition times.
[0061] For example, among the recognition results output by the "Xiao Ai" speech recognition model, 10,000 recognition results ranked top in the corresponding recognition times are used as the multiple candidate recognition texts obtained.
[0062] Step S120 , screening multiple candidate recognition texts to obtain a first target recognition text that is the same as the text in the preset corpus, where the text in the preset corpus is the accurately recognized text corresponding to the speech data to be recognized.
[0063] In the disclosed embodiment, a corpus, namely a preset corpus, may be pre-set to store accurately recognized texts corresponding to each speech data to be recognized, wherein the speech data to be recognized may be input speech data of a preset speech recognition model.
[0064] Optionally, the text in the preset corpus may be the accurately recognized text corresponding to the manual annotation. Optionally, the text in the preset corpus may also be the accurately recognized text output by the model for correcting the speech recognition result.
[0065] In the disclosed embodiment, after obtaining multiple candidate recognition texts, the multiple candidate recognition texts can be screened by determining whether each candidate recognition text exists in a preset corpus to obtain a first target recognition text that is the same as the text in the preset corpus.
[0066] That is, if there is a candidate recognition text in the preset corpus, the candidate recognition text can be determined as the first target recognition text.
[0067] Step S130, for any first target recognition text among the first target recognition texts, based on the first recognition number corresponding to the first target recognition text and the second recognition number corresponding to the first target recognition text in the time period before the first historical time period, obtain the recognition number change parameter corresponding to the first target recognition text.
[0068] Among them, the second recognition count corresponding to a first target recognition text represents the number of times the preset speech recognition model outputs the first target recognition text in a time period before the first historical time period.
[0069] It can be understood that after the above steps, multiple first target recognition texts may be obtained. At this time, for any first target recognition text among these multiple first target recognition texts, the recognition number change parameter corresponding to the first target recognition text can be obtained based on the first recognition number corresponding to the first target recognition text and the second recognition number corresponding to the first target recognition text.
[0070] The recognition times change parameter corresponding to the first target recognition text may reflect the changes in the number of times the first target recognition text has been recognized in history, for example, whether the first target recognition text has been steadily increasing or decreasing, explosively increasing or decreasing, etc. in history.
[0071] Optionally, the identification number change parameter may include at least one of an absolute change value, a relative burst value, and a frequency mutation value.
[0072] Step S140 , determining the recognition text to be annotated from each first target recognition text based on the recognition times change parameter corresponding to each first target recognition text.
[0073] The recognized text to be annotated can be understood as recognized text that needs to be annotated additionally, which may be recognized correctly or incorrectly, and the incorrect recognized text can be annotated with the correct text.
[0074] In the disclosed embodiment, after obtaining the recognition times change parameters corresponding to each first target recognition text, the recognition text to be annotated can be determined from each first target recognition text based on the recognition times change parameters corresponding to each first target recognition text.
[0075] In some embodiments, after determining the recognized texts to be annotated from each first target recognized text, each of the recognized texts to be annotated can be annotated. For example, the annotation can be performed manually to mark whether the recognized texts to be annotated are correct or incorrect, and the incorrect recognized texts to be annotated are annotated with correct texts.
[0076] The method for screening speech recognition results provided by the embodiment of the present disclosure is adopted. By obtaining multiple candidate recognition texts output by a preset speech recognition model in a first historical time period and the first recognition times corresponding to the multiple candidate recognition texts, the multiple candidate recognition texts are screened to obtain the first target recognition text that is the same as the text in the preset corpus. For any first target recognition text in each first target recognition text, based on the first recognition times corresponding to the first target recognition text and the second recognition times corresponding to the first target recognition text in the time period before the first historical time period, the recognition times change parameter corresponding to the first target recognition text is obtained. Based on the recognition times change parameters corresponding to each first target recognition text, the recognition text to be annotated is determined from each first target recognition text. The recognition text to be annotated is obtained by using the recognition times change parameters corresponding to the first target recognition text. On the one hand, the number of annotations for subsequent speech recognition results can be reduced, for example, the number of subsequent manual annotations can be reduced, saving human resources. On the other hand, since the recognition times of the accurately recognized recognition text usually change stably, if it is not stable, recognition errors may occur. Therefore, the recognition text to be annotated is determined from each first target recognition text by using the recognition times change parameters of the accurately recognized first target recognition text, so that the recognition text to be annotated can be accurately located.
[0077] In some implementations, the time period before the first historical time period includes a second historical time period before the first historical time period and adjacent to the first historical time period and having the same time length as the first historical time period. For example, the first historical time period is yesterday, and the second historical time period is the day before yesterday.
[0078] In this case, in step S130, obtaining the recognition times change parameter corresponding to the first target recognition text based on the first recognition times corresponding to the first target recognition text and the second recognition times corresponding to the first target recognition text may include the following steps:
[0079] Based on the first recognition times corresponding to the first target recognition text and the second recognition times corresponding to the first target recognition text in the second historical time period, a recognition times change parameter corresponding to the first target recognition text is obtained.
[0080] In the embodiment of the present disclosure, for any one of the first target recognition texts, the first recognition number and the second recognition number corresponding to the first target recognition text can be obtained, and then, based on the first recognition number and the second recognition number corresponding to the first target recognition text, the recognition number change parameter corresponding to the first target recognition text can be obtained.
[0081] Exemplarily, the first historical time period is yesterday, and the second historical time period is the day before yesterday, that is, the day before yesterday. At this time, the recognition number change parameter corresponding to the first target recognition text can be obtained based on the recognition number of the first target recognition text yesterday and the recognition number of the first target recognition text the day before yesterday.
[0082] In some implementations, obtaining a recognition times change parameter corresponding to the first target recognition text based on a first recognition times corresponding to the first target recognition text and a second recognition times corresponding to the first target recognition text in a second historical time period may include the following steps:
[0083] Determine a difference between a first recognition count corresponding to the first target recognition text and a second recognition count corresponding to the first target recognition text in a second historical time period;
[0084] Determine the difference as the recognition times change parameter corresponding to the first target recognition text, and / or determine the ratio of the difference to the second recognition times corresponding to the first target recognition text in the second historical time period as the recognition times change parameter corresponding to the first target recognition text;
[0085] Accordingly, in step S140, based on the recognition times change parameters corresponding to the first target recognition texts, determining the recognition texts to be annotated from the first target recognition texts may include the following steps:
[0086] From each first target recognition text, a preset number of first target recognition texts ranked higher in the corresponding recognition number change parameter are obtained as recognition texts to be annotated, and / or a preset number of first target recognition texts ranked lower in the corresponding recognition number change parameter are obtained as recognition texts to be annotated.
[0087] In an embodiment of the present disclosure, after determining the first recognition number corresponding to the first target recognition text and the second recognition number corresponding to the first target recognition text, the difference between the first recognition number corresponding to the first target recognition text and the corresponding second recognition number can be calculated.
[0088] Here, after calculating the difference between the first recognition times corresponding to the first target recognition text and the corresponding second recognition times, the recognition times change parameter corresponding to the first target recognition text can be obtained in different ways.
[0089] Optionally, the difference may be determined as a recognition times change parameter corresponding to the first target recognition text. The recognition times change parameter calculated at this time may also be referred to as an absolute change value. If the absolute change value is greater than 0, the corresponding first target recognition text may be referred to as an absolute increase text. If the absolute change value is less than 0, the corresponding first target recognition text may be referred to as an absolute decrease text.
[0090] Optionally, the ratio of the difference to the second recognition times corresponding to the first target recognition text in the second historical time period can be determined as the recognition times change parameter corresponding to the first target recognition text. The recognition times change parameter calculated at this time can also be called a relative burst value. If the relative burst value is greater than 0, the corresponding first target recognition text can be called a relatively high-burst text. If the relative burst value is less than 0, the corresponding first target recognition text can be called a relatively low-burst text.
[0091] Optionally, the difference, and the ratio of the difference to the second recognition number corresponding to the first target recognition text in the second historical time period, are both determined as recognition number change parameters corresponding to the first target recognition text.
[0092] In the embodiment of the present disclosure, when the difference is determined as the recognition number change parameter corresponding to the first target recognition text, and / or the ratio of the difference to the second recognition number corresponding to the first target recognition text in the second historical time period is determined as the recognition number change parameter corresponding to the first target recognition text, further, based on the recognition number change parameters corresponding to each first target recognition text, there can be a variety of different ways to determine the recognition text to be annotated from each first target recognition text.
[0093] Optionally, a preset number of first target recognition texts with corresponding recognition times change parameters being ranked top may be selected from among the first target recognition texts as the recognition texts to be annotated.
[0094] Optionally, a preset number of first target recognition texts whose corresponding recognition times change parameters are ranked low can be selected from among the first target recognition texts as the recognition texts to be annotated.
[0095] Optionally, from each first target recognition text, a preset number of first target recognition texts whose corresponding recognition number change parameters are ranked higher, and a preset number of first target recognition texts whose corresponding recognition number change parameters are ranked lower can be used as recognition texts to be marked.
[0096] That is, in the embodiment of the present disclosure, the text to be annotated determined from each first target recognition text may include at least one of the following:
[0097] After calculating the first recognition times corresponding to each first target recognition text and the difference between the corresponding second recognition times, a preset number of first target recognition texts with the top corresponding differences are selected from the differences corresponding to each first target recognition text as the texts to be annotated.
[0098] A preset number of first target recognition texts with lower corresponding difference values may be selected from the difference values respectively corresponding to the first target recognition texts as the texts to be annotated.
[0099] After calculating the ratio of the difference corresponding to each first target recognition text and the second recognition times corresponding to each of them, a preset number of first target recognition texts with the top corresponding ratios can be selected from the ratios corresponding to each first target recognition text as the texts to be marked.
[0100] From the ratios corresponding to the first target recognition texts, a preset number of first target recognition texts with lower corresponding ratios can be selected as the texts to be annotated.
[0101] It should be noted that, considering that the second recognition number may be 0, at this time, the sum of the second recognition number and the preset value can be obtained, and then the ratio of the difference and the sum can be determined as the recognition number change parameter corresponding to the first target recognition text. For example, the preset value can be 1.
[0102] In addition, in some embodiments, the first historical time period corresponds to a preset duration, and the time period before the first historical time period includes a second historical time period before the first historical time period and adjacent to the first historical time period and having a preset duration, and includes a third historical time period before the first historical time period and adjacent to the first historical time period and having a duration greater than the preset duration. For example, the preset duration can be one day, one week, one month, etc., determined according to actual needs.
[0103] For example, assuming the preset duration is one day, the first historical time period is yesterday, then the second historical time period is the day before yesterday, and the third historical time period can be the previous five, ten, thirty days, etc. days including yesterday, or it can be all time periods in history including yesterday, determined according to actual needs.
[0104] In this case, in step S130, obtaining the recognition times change parameter corresponding to the first target recognition text based on the first recognition times corresponding to the first target recognition text and the second recognition times corresponding to the first target recognition text may include the following steps:
[0105] Based on the first recognition number corresponding to the first target recognition text, the second recognition number corresponding to the first target recognition text in the second historical time period, and the third recognition number corresponding to the first target recognition text in the third historical time period, the recognition number change parameter corresponding to the first target recognition text is obtained.
[0106] In the embodiment of the present disclosure, for any one of the first target recognition texts, the first recognition number, the second recognition number and the third recognition number corresponding to the first target recognition text can also be obtained. Then, based on the first recognition number, the second recognition number and the third recognition number corresponding to the first target recognition text, the recognition number change parameter corresponding to the first target recognition text can be obtained.
[0107] In some implementations, obtaining a recognition times change parameter corresponding to the first target recognition text based on a first recognition times corresponding to the first target recognition text, a second recognition times corresponding to the first target recognition text in a second historical time period, and a third recognition times corresponding to the first target recognition text in a third historical time period may include the following steps:
[0108] The product of the first value and the second value is determined as the recognition number change parameter corresponding to the first target recognition text, the first value is the ratio of the first recognition number corresponding to the first target recognition text to the corresponding third recognition number, and the second value is obtained based on the second recognition number corresponding to the first target recognition text and a preset logarithmic function;
[0109] Accordingly, in step S140, based on the recognition times change parameters corresponding to the first target recognition texts, determining the recognition texts to be annotated from the first target recognition texts may include the following steps:
[0110] From each first target recognition text, a preset number of first target recognition texts whose corresponding recognition times change parameters are ranked top are obtained as recognition texts to be annotated.
[0111] In the disclosed embodiment, the ratio of the first recognition times and the corresponding third recognition times corresponding to each first target recognition text, i.e., the first value, can be obtained. In addition, the second value corresponding to each first target recognition text can be obtained based on the second recognition times corresponding to each first target recognition text and the preset logarithmic function. Then, the product of the first value and the second value corresponding to each first target recognition text can be determined as the recognition times change parameter corresponding to each first target recognition text. The recognition times change parameter calculated at this time can also be called a frequency mutation value.
[0112] In some implementations, the preset logarithmic function may be, for example, a log function, or an lg function, or an ln function.
[0113] In the disclosed embodiment, after obtaining the recognition times change parameter corresponding to each first target recognition text, a preset number of first target recognition texts ranked high in the corresponding recognition times change parameter can be obtained from each first target recognition text as recognition texts to be annotated.
[0114] In addition, in some implementations, the method of the embodiment of the present disclosure may further include the following steps:
[0115] According to the first recognition times corresponding to the plurality of candidate recognition texts, a preset number of third target recognition texts whose corresponding first recognition times are ranked top are determined from the plurality of candidate recognition texts;
[0116] According to the third target recognition text, the recognition text to be processed is determined.
[0117] Among them, the recognition text to be processed can be understood as the recognition text that is the focus of attention.
[0118] In the embodiment of the present disclosure, the first recognition times corresponding to each of a plurality of candidate recognition texts can be obtained, and then a preset number of third target recognition texts whose corresponding first recognition times are ranked high can be determined from the plurality of candidate recognition texts, and then the recognition text to be processed can be determined based on the third target recognition texts.
[0119] Continuing with the above example, assuming that among the recognition results output by the "Xiao Ai" speech recognition model, the top 10,000 recognition results with the corresponding recognition times are used as the multiple candidate recognition texts obtained, a preset number of third target recognition texts with the corresponding first recognition times can be selected from the 10,000 candidate recognition texts, and the third target recognition texts are also used as the texts to be processed.
[0120] For example, for the "Xiao Ai Tongxue" speech recognition model, the recognition text "Xiao Ai Tongxue" is a high-frequency text requested by online speaker users. If this text is not in the third target recognition text in the data within the first historical time period, it needs to be paid special attention.
[0121] In the disclosed embodiment, the recognition text to be processed is determined based on the third target recognition text, which can stably monitor the high-frequency data in the first historical time period and help quickly discover anomalies.
[0122] In some implementations, the method of the embodiment of the present disclosure may further include the following steps:
[0123] A plurality of candidate recognition texts are screened to obtain a second target recognition text that is different from the text in the preset corpus, and the second target recognition text is used for manual recognition and annotation.
[0124] In the embodiments of the present disclosure, considering that the text in the preset corpus is the accurately recognized text corresponding to the speech data to be recognized, if the candidate recognition text is not in the preset corpus, it may be an incorrectly recognized text. Therefore, after obtaining multiple candidate recognition texts, the multiple candidate recognition texts can be screened to obtain a second target recognition text that is different from the text in the preset corpus, wherein the obtained second target recognition text can be used for manual recognition and annotation, which can reduce the workload of manual annotation compared to manually annotating all candidate recognition texts.
[0125] In some embodiments, considering the situation of insufficient manpower, a preset number of second target recognition texts with the highest number of corresponding recognition times can be selected from the second target recognition texts for manual screening. In the case of insufficient manpower, selecting the second target recognition texts with a large number of recognition times for manual annotation can improve the efficiency of manual annotation while ensuring a high correction effect.
[0126] In some implementations, the method of the embodiment of the present disclosure may further include the following steps:
[0127] Acquire the accurately recognized text of the voice data corresponding to the second target recognized text;
[0128] Add the accurately recognized text to the preset corpus.
[0129] In the embodiments of the present disclosure, combined with the foregoing content, it can be known that after obtaining the second target recognition text, the second target recognition text can be manually recognized and annotated, so as to obtain the accurately recognized text of the voice data corresponding to the second target recognition text. Then, the accurately recognized text can be further added to the preset corpus to continuously update the preset corpus. By continuously updating the preset corpus, the preset corpus can be continuously enriched and expanded, so as to filter the candidate recognition texts collected subsequently, reduce the amount of voice data that needs to be manually annotated, and save human resources.
[0130] It should be noted that the method in the embodiment of the present disclosure can be continuously executed in a normalized manner over time, thereby continuously enriching the preset corpus and continuously mining the text to be annotated and recognized in the process.
[0131] It should be noted that, in the embodiments of the present disclosure, the numerical values of the preset numbers can be set according to actual needs, and the various preset numbers can be the same or different.
[0132] Next, combine Figure 2 The flowchart shown uses a specific example to illustrate the method for screening the speech recognition results of the embodiment of the present disclosure.
[0133] like Figure 2 As shown, the electronic device can obtain the "Xiao Ai" voice to recognize the user's voice within one day yesterday, sort the recognized texts according to the number of recognitions of each recognized text, select the top 10,000 recognized texts as candidate recognized texts, and select the top 100 candidate recognized texts with the highest number of recognitions from the 10,000 candidate recognized texts as the third target recognized texts, which can also be called the full TOP texts. For the convenience of description, yesterday will be recorded as the first day in the following.
[0134] In addition, the currently accumulated preset corpus can be obtained, and according to whether the candidate recognition text exists in the preset corpus, 10,000 candidate recognition texts are divided into first target recognition texts in the preset corpus and second target recognition texts not in the preset corpus.
[0135] For the screened second target recognition texts, 100 second target recognition texts ranked top in recognition times can be selected as TOP texts not in the preset corpus for manual annotation.
[0136] For the first target recognition texts obtained by screening, the recognition times of each first target text in the day before yesterday and the recognition times of each first target text in the previous 7 days including yesterday can be obtained. For the convenience of description, the day before yesterday is recorded as the second day, and the previous 7 days including yesterday are recorded as the last week.
[0137] Subsequently, various recognition times change parameters can be calculated based on the obtained recognition times. Taking various recognition times change parameters corresponding to a first target text as an example, the calculation method is as follows:
[0138] Absolute change value = number of recognitions on the first day - number of recognitions on the second day;
[0139] Relative burst value = [(number of recognitions on the first day - number of recognitions on the second day) / (number of recognitions on the second day + 1)]%;
[0140] Frequency mutation value = (number of recognitions on the first day / number of recognitions last week) * log (number of recognitions on the second day);
[0141] Through the above calculation process, the absolute change value, relative burst value and frequency mutation value corresponding to each first target text can be obtained. Then, the absolute change value can be sorted from large to small, and the top 100 first target recognition texts can be selected as the absolutely increased TOP texts, and the bottom 100 first target recognition texts can be selected as the absolutely decreased TOP texts.
[0142] According to the relative burst value sorting from large to small, the top 100 first target recognition texts are selected as the relatively high-burst TOP texts, and the bottom 100 first target recognition texts are selected as the relatively low-burst TOP texts.
[0143] Sort by frequency mutation value from large to small, and select the top 100 first target recognition texts as the frequency mutation TOP texts.
[0144] After the above process, 100 full TOP texts, 100 TOP texts not in the preset corpus, 100 absolutely increased TOP texts, 100 absolutely decreased TOP texts, 100 relatively high TOP texts, 100 relatively low TOP texts, and 100 frequency mutation TOP texts can be obtained. Among them, the recognition number change parameter can include absolute change value, relative burst value, and frequency mutation value.
[0145] Finally, the 700 texts obtained can be manually annotated. It should be noted that there may be repeated texts in the 700 texts obtained.
[0146] In addition, in some embodiments, Figure 3 As shown in the figure, the various TOP texts obtained can be visualized in the interface. It should be noted that due to space reasons, Figure 3 The top 12 of each category of TOP texts are shown in the figure.
[0147] For example, Figure 3The data for November 8, 2022 is shown in Figure 3 As shown, the full TOP text displayed corresponds to the number of recognitions on November 8, 2022, the TOP text not in the preset corpus displayed corresponds to the number of recognitions on November 8, 2022, the TOP text with absolute increase displayed corresponds to the absolute change value on November 8, 2022 (a positive integer), the TOP text with absolute decrease displayed corresponds to the absolute change value on November 8, 2022 (a negative integer), the TOP text with relative sudden increase displayed corresponds to the relative sudden value on November 8, 2022 (a positive percentage), the TOP text with relative sudden decrease displayed corresponds to the relative sudden value on November 8, 2022 (a negative percentage), and the TOP text with frequency mutation displayed corresponds to the frequency mutation value on November 8, 2022.
[0148] Figure 4 is a block diagram of a device for filtering speech recognition results according to an exemplary embodiment. Figure 4 The speech recognition result screening device 400 is applied to an electronic device, and the speech recognition result screening device 400 includes:
[0149] The acquisition module 410 is configured to acquire a plurality of candidate recognition texts output by a preset speech recognition model in a first historical time period, and first recognition times corresponding to the plurality of candidate recognition texts, respectively. The first recognition time corresponding to a candidate recognition text represents the number of times the preset speech recognition model outputs the candidate recognition text in the first historical time period;
[0150] A screening module 420 is configured to screen the plurality of candidate recognition texts to obtain a first target recognition text that is the same as a text in a preset corpus, wherein the text in the preset corpus is an accurately recognized text corresponding to the speech data to be recognized;
[0151] The calculation module 430 is configured to obtain, for any first target recognition text among the first target recognition texts, a recognition number change parameter corresponding to the first target recognition text based on the first recognition number corresponding to the first target recognition text and the second recognition number corresponding to the first target recognition text, wherein the second recognition number corresponding to the first target recognition text represents the number of times the preset speech recognition model outputs the first target recognition text in a time period before the first historical time period;
[0152] The determination module 440 is configured to determine the recognized text to be annotated from each of the first target recognized texts based on the recognition times change parameter corresponding to each of the first target recognized texts.
[0153] Optionally, the time period before the first historical time period includes a second historical time period before the first historical time period and adjacent to the first historical time period and having the same time length, and the calculation module 430 includes:
[0154] The first calculation submodule is configured to obtain a recognition times change parameter corresponding to the first target recognition text based on the first recognition times corresponding to the first target recognition text and the second recognition times corresponding to the first target recognition text in the second historical time period.
[0155] Optionally, the calculation submodule includes:
[0156] A first calculation unit is configured to determine a difference between the first recognition number corresponding to the first target recognition text and the second recognition number corresponding to the first target recognition text in the second historical time period;
[0157] The second calculation unit is configured to determine the difference as the recognition number change parameter corresponding to the first target recognition text, and / or determine the ratio of the difference to the second recognition number corresponding to the first target recognition text in the second historical time period as the recognition number change parameter corresponding to the first target recognition text.
[0158] Correspondingly, the determination module 440 is also configured to obtain, from each of the first target recognition texts, a preset number of first target recognition texts that are ranked higher in the corresponding recognition number change parameter as the recognition texts to be annotated, and / or obtain a preset number of first target recognition texts that are ranked lower in the corresponding recognition number change parameter as the recognition texts to be annotated.
[0159] Optionally, the first historical time period corresponds to a preset duration, the time period before the first historical time period includes a second historical time period before the first historical time period and adjacent to the first historical time period and having the preset duration, and includes a third historical time period before the first historical time period and adjacent to the first historical time period and having a duration greater than the preset duration, and the calculation module 430 includes:
[0160] The second calculation submodule is configured to obtain the recognition number change parameter corresponding to the first target recognition text based on the first recognition number corresponding to the first target recognition text, the second recognition number corresponding to the first target recognition text in the second historical time period, and the third recognition number corresponding to the first target recognition text in the third historical time period.
[0161] Optionally, the second calculation submodule includes:
[0162] The third calculation unit is configured to determine the product of the first numerical value and the second numerical value as the recognition number change parameter corresponding to the first target recognition text, wherein the first numerical value is the ratio of the first recognition number corresponding to the first target recognition text to the corresponding third recognition number, and the second numerical value is obtained based on the second recognition number corresponding to the first target recognition text and a preset logarithmic function.
[0163] Correspondingly, the determination module 440 is further configured to obtain, from each of the first target recognition texts, a preset number of first target recognition texts that are ranked top in terms of the corresponding recognition times change parameter as the recognition texts to be annotated.
[0164] Optionally, the apparatus 400 further includes:
[0165] A third target recognition text determination module is configured to determine, according to the first recognition times corresponding to the plurality of candidate recognition texts, a preset number of third target recognition texts whose corresponding first recognition times are ranked top from the plurality of candidate recognition texts;
[0166] The module for determining the recognized text to be processed is configured to determine the recognized text to be processed according to the third target recognized text.
[0167] Optionally, the apparatus 400 further includes:
[0168] The second target recognition text screening module is configured to screen the multiple candidate recognition texts to obtain a second target recognition text that is different from the text in the preset corpus, and the second target recognition text is used for manual recognition and annotation.
[0169] Optionally, the apparatus 400 further includes:
[0170] An accurately recognized text acquisition module is configured to acquire an accurately recognized text of the voice data corresponding to the second target recognized text;
[0171] The adding module is configured to add the accurately recognized text to the preset corpus.
[0172] Regarding the device for filtering speech recognition results in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.
[0173] The present disclosure also provides a computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, implement the steps of the method for screening speech recognition results provided by the present disclosure.
[0174] Figure 51 is a block diagram of an electronic device for a method for screening speech recognition results according to an exemplary embodiment. The electronic device 500 may be, for example, a mobile phone, a computer, a tablet device, or the like.
[0175] Reference Figure 5 , the electronic device 500 may include one or more of the following components: a processing component 502 , a memory 504 , a power component 506 , a multimedia component 508 , an audio component 510 , an input / output interface 512 , a sensor component 514 , and a communication component 516 .
[0176] The processing component 502 generally controls the overall operation of the electronic device 500, such as operations associated with display, phone calls, data communications, camera operations, and recording operations. The processing component 502 may include one or more processors 520 to execute instructions to complete all or part of the steps of the above-mentioned method. In addition, the processing component 502 may include one or more modules to facilitate the interaction between the processing component 502 and other components. For example, the processing component 502 may include a multimedia module to facilitate the interaction between the multimedia component 508 and the processing component 502.
[0177] The memory 504 is configured to store various types of data to support operations on the electronic device 500. Examples of such data include instructions for any application or method operating on the electronic device 500, contact data, phone book data, messages, pictures, videos, etc. The memory 504 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.
[0178] The power supply component 506 provides power to the various components of the electronic device 500. The power supply component 506 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the electronic device 500.
[0179] The multimedia component 508 includes a screen that provides an output interface between the electronic device 500 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundaries of the touch or slide action, but also detect the duration and pressure associated with the touch or slide operation. In some embodiments, the multimedia component 508 includes a front camera and / or a rear camera. When the electronic device 500 is in an operating mode, such as a shooting mode or a video mode, the front camera and / or the rear camera may receive external multimedia data. Each front camera and rear camera may be a fixed optical lens system or have a focal length and optical zoom capability.
[0180] The audio component 510 is configured to output and / or input audio signals. For example, the audio component 510 includes a microphone (MIC), and when the electronic device 500 is in an operating mode, such as a call mode, a recording mode, and a voice recognition mode, the microphone is configured to receive an external audio signal. The received audio signal can be further stored in the memory 504 or sent via the communication component 516. In some embodiments, the audio component 510 also includes a speaker for outputting audio signals.
[0181] The input / output interface 512 provides an interface between the processing component 502 and the peripheral interface modules, which may be keyboards, click wheels, buttons, etc. These buttons may include but are not limited to: a home button, a volume button, a start button, and a lock button.
[0182] The sensor assembly 514 includes one or more sensors for providing various aspects of status assessment for the electronic device 500. For example, the sensor assembly 514 can detect the open / closed state of the electronic device 500, the relative positioning of components, such as the display and keypad of the electronic device 500, and the sensor assembly 514 can also detect the position change of the electronic device 500 or a component of the electronic device 500, the presence or absence of user contact with the electronic device 500, the orientation or acceleration / deceleration of the electronic device 500, and the temperature change of the electronic device 500. The sensor assembly 514 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. The sensor assembly 514 may also include an optical sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor assembly 514 may also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.
[0183] The communication component 516 is configured to facilitate wired or wireless communication between the electronic device 500 and other devices. The electronic device 500 can access a wireless network based on a communication standard, such as WiFi, 2G or 3G, or a combination thereof. In an exemplary embodiment, the communication component 516 receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 516 also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.
[0184] In an exemplary embodiment, the electronic device 500 can be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the above-mentioned method for screening speech recognition results.
[0185] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory 504 including instructions, and the instructions can be executed by a processor 520 of an electronic device 500 to perform the above method. For example, the non-transitory computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.
[0186] In another exemplary embodiment, a computer program product is also provided. The computer program product includes a computer program that can be executed by a programmable device. The computer program has a code portion for executing the above-mentioned method for screening speech recognition results when executed by the programmable device.
[0187] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the present disclosure. The present disclosure is intended to cover any variations, uses or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The description and examples are to be considered as exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.
[0188] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.
Claims
1. A method for screening speech recognition results, characterized in that: The method comprises: Acquire multiple candidate recognition texts output by a preset speech recognition model in a first historical time period, and first recognition times corresponding to the multiple candidate recognition texts respectively. The first recognition time corresponding to a candidate recognition text represents the number of times the preset speech recognition model outputs the candidate recognition text in the first historical time period; Screening the plurality of candidate recognition texts to obtain a first target recognition text that is identical to a text in a preset corpus, wherein the text in the preset corpus is an accurately recognized text corresponding to the speech data to be recognized; For any first target recognition text among the first target recognition texts, based on the first recognition number corresponding to the first target recognition text and the second recognition number corresponding to the first target recognition text, a recognition number change parameter corresponding to the first target recognition text is obtained, wherein the second recognition number corresponding to the first target recognition text represents the number of times the preset speech recognition model outputs the first target recognition text in a time period before the first historical time period; Based on the recognition times variation parameters corresponding to the first target recognition texts, the recognition texts to be annotated are determined from the first target recognition texts.
2. The method according to claim 1, characterized in that The time period before the first historical time period includes a second historical time period before the first historical time period and adjacent to the first historical time period and having the same time length, The obtaining, based on the first recognition number corresponding to the first target recognition text and the second recognition number corresponding to the first target recognition text, a recognition number change parameter corresponding to the first target recognition text includes: Based on the first recognition times corresponding to the first target recognition text and the second recognition times corresponding to the first target recognition text in the second historical time period, a recognition times change parameter corresponding to the first target recognition text is obtained.
3. The method according to claim 2, characterized in that The obtaining of a recognition times change parameter corresponding to the first target recognition text based on the first recognition times corresponding to the first target recognition text and the second recognition times corresponding to the first target recognition text in the second historical time period includes: Determine a difference between the first recognition count corresponding to the first target recognition text and the second recognition count corresponding to the first target recognition text in the second historical time period; Determine the difference as the recognition times change parameter corresponding to the first target recognition text, and / or determine the ratio of the difference to the second recognition times corresponding to the first target recognition text in the second historical time period as the recognition times change parameter corresponding to the first target recognition text; The determining the to-be-annotated recognition text from each of the first target recognition texts based on the recognition number change parameters corresponding to each of the first target recognition texts includes: From each of the first target recognition texts, a preset number of first target recognition texts ranked higher in the corresponding recognition number change parameter are obtained as the recognition texts to be annotated, and / or a preset number of first target recognition texts ranked lower in the corresponding recognition number change parameter are obtained as the recognition texts to be annotated.
4. The method according to claim 1, characterized in that: The first historical time period corresponds to a preset duration, the time period before the first historical time period includes a second historical time period before the first historical time period and adjacent to the first historical time period and having the preset duration, and includes a third historical time period before the first historical time period and adjacent to the first historical time period and having a duration greater than the preset duration, and the first recognition count corresponding to the first target recognition text and the second recognition count corresponding to the first target recognition text are used to obtain a recognition count change parameter corresponding to the first target recognition text, including: Based on the first recognition times corresponding to the first target recognition text, the second recognition times corresponding to the first target recognition text in the second historical time period, and the third recognition times corresponding to the first target recognition text in the third historical time period, the recognition times change parameter corresponding to the first target recognition text is obtained.
5. The method according to claim 4, characterized in that The obtaining of a recognition number change parameter corresponding to the first target recognition text based on the first recognition number corresponding to the first target recognition text, the second recognition number corresponding to the first target recognition text in the second historical time period, and the third recognition number corresponding to the first target recognition text in the third historical time period includes: The product of a first value and a second value is determined as a recognition times change parameter corresponding to the first target recognition text, wherein the first value is a ratio of the first recognition times corresponding to the first target recognition text to the corresponding third recognition times, and the second value is obtained based on the second recognition times corresponding to the first target recognition text and a preset logarithmic function; The determining the to-be-annotated recognition text from each of the first target recognition texts based on the recognition number change parameters corresponding to each of the first target recognition texts includes: From each of the first target recognition texts, a preset number of first target recognition texts that are ranked top in the corresponding recognition times change parameter are obtained as the recognition texts to be annotated.
6. The method according to claim 1, characterized in that The method further comprises: According to the first recognition times corresponding to the plurality of candidate recognition texts, respectively, determining a preset number of third target recognition texts whose corresponding first recognition times are ranked top from the plurality of candidate recognition texts; According to the third target recognition text, a recognition text to be processed is determined.
7. The method according to claim 1, characterized in that The method further comprises: The multiple candidate recognition texts are screened to obtain a second target recognition text that is different from the text in the preset corpus, and the second target recognition text is used for manual recognition and annotation.
8. The method according to claim 7, characterized in that The method further comprises: Acquire the accurately recognized text of the speech data corresponding to the second target recognized text; The accurately recognized text is added to the preset corpus.
9. A device for screening speech recognition results, characterized in that: The device comprises: An acquisition module is configured to acquire a plurality of candidate recognition texts output by a preset speech recognition model in a first historical time period, and first recognition times corresponding to the plurality of candidate recognition texts, respectively. The first recognition time corresponding to a candidate recognition text represents the number of times the preset speech recognition model outputs the candidate recognition text in the first historical time period; A screening module is configured to screen the plurality of candidate recognition texts to obtain a first target recognition text that is the same as a text in a preset corpus, wherein the text in the preset corpus is an accurately recognized text corresponding to the speech data to be recognized; A calculation module is configured to obtain, for any first target recognition text among the first target recognition texts, a recognition number change parameter corresponding to the first target recognition text based on the first recognition number corresponding to the first target recognition text and the second recognition number corresponding to the first target recognition text, wherein the second recognition number corresponding to the first target recognition text represents the number of times the preset speech recognition model outputs the first target recognition text in a time period before the first historical time period; The determination module is configured to determine the recognition text to be marked from each of the first target recognition texts based on the recognition number change parameters corresponding to each of the first target recognition texts.
10. An electronic device, characterized in that: The electronic device comprises: processor; a memory for storing processor-executable instructions; The processor is configured to implement the steps of any one of the methods of claims 1 to 8 when executing.
11. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the program instructions are executed by a processor, the steps of the method described in any one of claims 1 to 8 are implemented.
Citation Information
Patent Citations
Voice recognition model generation method and device, storage medium and electronic equipment
CN108847222A
Text recognition method and device, electronic equipment and storage medium
CN115291791A
Voice user recognition method and device, electronic equipment and storage medium
CN115662443A
Data collection method and device, text recognition method and device, equipment and storage medium
CN115796164A
Silicon wafer and epitaxial silicon wafer
KR1020230065174A
