Simultaneous transmission quality evaluation method, related device, equipment and storage medium

Abstract

The invention discloses a simultaneous transmission quality evaluation method, a related device, equipment and a storage medium, and the method comprises the steps: carrying out the segmentation of a simultaneous transmission text based on a target voice, and obtaining a plurality of sub-texts; acquiring word brushing data of sub-voices corresponding to the sub-texts in the target voice; wherein the word brushing data of the sub-voices comprises a plurality of texts from the first word output to the gradual correction until the sub-voices are finally translated into the sub-texts in the simultaneous transmission process of the sub-voices; based on the word brushing data of the sub-voices, simultaneous transmission quality scores of the sub-voices are obtained through analysis; and based on the simultaneous transmission quality scores of the sub-voices, performing fusion to obtain a target score representing the simultaneous transmission quality of the target voice. According to the scheme, the accuracy of simultaneous transmission quality evaluation can be improved in an application scene of streaming simultaneous transmission.

Description

Technical Field

[0001] The present application relates to the technical field of natural language processing, and in particular to a simultaneous interpretation quality evaluation method and related devices, equipment and storage media. Background Art

[0002] Thanks to the continuous development of machine learning, the application of big models to simultaneous interpretation has made considerable progress. Unlike traditional simultaneous interpretation, big model simultaneous interpretation requires that the corresponding translation results are directly generated in a streaming manner while the user inputs the voice, without the need for recognition and then translation. Therefore, it is particularly important to evaluate the quality of simultaneous interpretation end-to-end.

[0003] However, the existing evaluation methods are usually applicable to traditional simultaneous interpretation. If applied to large-scale simultaneous interpretation, it will be difficult to truly reflect the quality of simultaneous interpretation. In view of this, how to improve the accuracy of simultaneous interpretation quality evaluation in the application scenario of streaming simultaneous interpretation has become an urgent problem to be solved. Summary of the invention

[0004] The main technical problem solved by the present application is to provide a simultaneous interpretation quality evaluation method and related devices, equipment and storage media, which can improve the accuracy of simultaneous interpretation quality evaluation in the application scenario of streaming simultaneous interpretation.

[0005] In order to solve the above technical problems, the first aspect of the present application provides a simultaneous interpretation quality evaluation method, including: segmenting the simultaneous interpretation text based on the target speech to obtain a plurality of sub-texts; obtaining the brush word data of the sub-speech corresponding to the sub-text in the target speech; wherein the brush word data of the sub-speech includes: a plurality of texts of the sub-speech from the first word output to the gradual correction until the final translation into the sub-text during the simultaneous interpretation process; based on the brush word data of the sub-speech, analyzing and obtaining the simultaneous interpretation quality score of the sub-speech; based on the simultaneous interpretation quality scores of each sub-speech, fusing and obtaining the target score representing the simultaneous interpretation quality of the target speech.

[0006] In order to solve the above-mentioned technical problems, the second aspect of the present application provides a simultaneous interpretation quality evaluation device, including: a text segmentation module, a data acquisition module, a quality analysis module and a scoring fusion module, the text segmentation module is used to segment the simultaneous interpretation text based on the target speech to obtain a plurality of sub-texts; the data acquisition module is used to obtain the brush word data of the sub-speech corresponding to the sub-text in the target speech; wherein the brush word data of the sub-speech includes: a plurality of texts of the sub-speech from the first word output to the gradual correction until the final translation into the sub-text during the simultaneous interpretation process; the quality analysis module is used to analyze the brush word data of the sub-speech to obtain the simultaneous interpretation quality score of the sub-speech based on the sub-speech; the scoring fusion module is used to fuse the simultaneous interpretation quality scores of each sub-speech to obtain a target score representing the simultaneous interpretation quality of the target speech.

[0007] In order to solve the above technical problems, the third aspect of the present application provides an electronic device, which at least includes a memory and a processor coupled to each other, wherein the memory at least stores program instructions, and the processor is used to execute the program instructions to implement the simultaneous interpretation quality assessment method in the above first aspect.

[0008] In order to solve the above technical problems, the fourth aspect of the present application provides a computer-readable storage medium storing program instructions that can be executed by a processor, and the program instructions are used to implement the simultaneous interpretation quality evaluation method of the first aspect.

[0009] The above scheme is based on segmenting the simultaneous interpretation text of the target speech to obtain several sub-texts, and obtaining the brush word data of the sub-speech corresponding to the sub-text in the target speech, and the brush word data of the sub-speech includes: several texts of the sub-speech from the first word to the gradual correction until the final translation into the sub-text during the simultaneous interpretation process, and then based on the brush word data of the sub-speech, the simultaneous interpretation quality score of the sub-speech is analyzed, and then based on the simultaneous interpretation quality scores of each sub-speech, the target score representing the simultaneous interpretation quality of the target speech is obtained by fusion. Therefore, on the one hand, by dividing the simultaneous interpretation text into several sub-texts, and then performing simultaneous interpretation on each sub-text corresponding to the sub-speech Compared with the overall quality evaluation of the target speech and its simultaneous interpretation text, the simultaneous interpretation quality evaluation and final score fusion can further refine the evaluation granularity, which helps to improve the accuracy of the simultaneous interpretation quality evaluation to a certain extent. On the other hand, when evaluating each sub-speech, the brush word data of the sub-speech is combined, and the brush word data includes the sub-speech from the first word to the gradual correction and finally the translation into a sub-text during the simultaneous interpretation process. Therefore, in the process of simultaneous interpretation quality evaluation of the sub-speech, attention can be paid to the simultaneous interpretation process, which helps to measure the simultaneous interpretation quality in the brush word process in the application scenario of streaming simultaneous interpretation. Therefore, the accuracy of simultaneous interpretation quality evaluation can be improved in the application scenario of streaming simultaneous interpretation. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is a flowchart of an embodiment of a simultaneous interpretation quality evaluation method of the present application; Figure 2a It is a schematic diagram of an embodiment of brush character data; Figure 2b is a schematic diagram of an embodiment of a simultaneous interpretation sequence; Figure 2c It is a schematic diagram of another embodiment of brush character data; Figure 2d It is a schematic diagram of the process of an embodiment of simultaneous interpretation quality evaluation of the present application; Figure 3 It is a schematic diagram of the framework of an embodiment of the simultaneous interpretation quality evaluation device of the present application; Figure 4 It is a schematic diagram of the framework of an embodiment of the electronic device of the present application; Figure 5 It is a schematic diagram of a framework of an embodiment of a computer-readable storage medium of the present application. DETAILED DESCRIPTION

[0011] The scheme of the embodiment of the present application is described in detail below in conjunction with the drawings of the specification.

[0012] In the following description, for the purpose of explanation rather than limitation, specific details such as specific system structures, interfaces, and technologies are provided to facilitate a thorough understanding of the present application.

[0013] The terms "system" and "network" are often used interchangeably in this article. The term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, the fragment " / " in this article generally indicates that the associated objects before and after are in an "or" relationship. In addition, "many" in this article means two or more than two.

[0014] See also Figure 1 , Figure 1 This is a flowchart of an embodiment of the simultaneous interpretation quality evaluation method of the present application. Specifically, it may include the following steps: Step S11: segmenting the simultaneous interpretation text based on the target speech to obtain a number of sub-texts.

[0015] In one implementation scenario, the simultaneous interpretation text can be obtained by interpreting the target speech by a simultaneous interpretation big model. It should be noted that the simultaneous interpretation big model may include but is not limited to: open source big models such as LLAMA, Bloom, etc., and may also include but is not limited to a big model obtained by fine-tuning the parameters of an open source big model through a specific corpus, or may also include but is not limited to a custom big model. The specific source of the simultaneous interpretation big model is not limited here.

[0016] In an implementation scenario, before segmenting the simultaneous interpretation text based on the target speech, the simultaneous interpretation text may be preprocessed, such as but not limited to: removing redundant spaces, special characters, etc., to ensure text quality.

[0017] In an implementation scenario, the simultaneous interpretation text can be segmented based on punctuation marks to obtain several sub-texts. For example, the simultaneous interpretation text can be segmented based on punctuation marks such as semicolons, commas, and periods to obtain several sub-texts. For example, taking the simultaneous interpretation text "Today's meeting discussed the progress of the project and put forward constructive suggestions; at the same time, the manager made a summary and arranged the next plan" as an example, when segmented based on punctuation marks, the following sub-texts can be obtained: "Today's meeting discussed the progress of the project", "put forward constructive suggestions", "at the same time", "the manager made a summary", "and arranged the next plan". Of course, the above example is only a possible example of text segmentation in actual application, and other possible situations will not be given one by one here.

[0018] In another implementation scenario, different from the aforementioned implementation, the simultaneous interpretation text can also be segmented based on semantics to obtain several sub-texts. For example, the simultaneous interpretation text can be segmented based on a semantic segmentation model to obtain several sub-texts. Exemplarily, still taking the simultaneous interpretation text "Today's meeting discussed the progress of the project and put forward constructive suggestions; at the same time, the manager made a summary and arranged the next plan" as an example, when segmented based on semantics, the following sub-texts can be obtained: "Today's meeting discussed the progress of the project", "put forward constructive suggestions", "At the same time, the manager made a summary", "and arranged the next plan". Of course, the above example is only a possible example of text segmentation in actual application, and other possible situations will not be given one by one here.

[0019] In another implementation scenario, different from the above implementation, the simultaneous interpretation text can also be segmented based on punctuation to obtain several first subtexts, and the simultaneous interpretation text can be segmented based on semantics to obtain several second subtexts, and then integrated based on several first subtexts and several second subtexts to obtain several subtexts. For example, in the integration process, if it is found that any first subtext belongs to a part of a second subtext, the first subtext can follow the segmentation method of the second subtext to which it belongs. On the contrary, if it is found that any second subtext belongs to a part of a first subtext, the second subtext can follow the segmentation method of the first subtext to which it belongs, so as to achieve integration of the two segmentation methods. Still taking the simultaneous interpretation text "Today's meeting discussed the progress of the project and put forward constructive suggestions; at the same time, the manager made a summary and arranged the next step" as an example, since the first subtext "at the same time" belongs to a part of the second subtext "at the same time, the manager made a summary", the first subtext "at the same time" can follow the segmentation method of the second subtext "at the same time, the manager made a summary". In addition, based on the above integration, the adjacent subtexts of each subtext can be further examined. If the adjacent subtext has logical coherence with it, the adjacent subtext can be further segmented into the same subtext as the subtext. Still taking the above example, after the above integration, the following subtexts can be obtained: "Today's meeting discussed the progress of the project", "Constructive opinions were put forward", "At the same time, the manager made a summary", "and arranged the next plan". By examining the logical coherence between adjacent subtexts, the subtext "Today's meeting discussed the progress of the project" and the subtext "Constructive opinions were put forward" can be integrated into a subtext "Today's meeting discussed the progress of the project and put forward constructive opinions", and the subtext "At the same time, the manager made a summary" and the subtext "And arranged the next plan" can be integrated into a subtext "At the same time, the manager made a summary and arranged the next plan". Of course, the above example is only a possible example of subtext integration, and other possible situations will not be given one by one here. The above method divides the simultaneous interpretation text based on punctuation to obtain several first sub-texts, and divides the simultaneous interpretation text based on semantics to obtain several second sub-texts, and then integrates the several first sub-texts and the several second sub-texts to obtain several sub-texts. This method can combine grammatical structure and semantic information to more accurately reflect the logical relationship and meaning of the sentence, making the result more natural and coherent.

[0020] Step S12: Obtain the brush character data of the sub-speech corresponding to the sub-text in the target speech.

[0021] In the embodiment of the present disclosure, the sub-speech word brushing data includes: a number of texts of the sub-speech from the initial word output to the gradual correction and final translation into the sub-text during the simultaneous interpretation process. Figure 2a , Figure 2aSchematic diagram of an embodiment of brush character data. Figure 2a As shown, the gray text indicates that the system directly translates the word because the speaker has not finished speaking. The subsequent results may change due to factors such as grammar and context. The black text indicates the definite translation result given by the system after comprehensively considering the context and grammar. The result will not change in the future. Figure 2a The above is only a possible example of word-brushing data, and the specific content of word-brushing data is not limited here. Taking the above example, we can obtain word-brushing data corresponding to the sub-speech of the sub-text "Today's meeting discussed the progress of the project and put forward constructive suggestions", and obtain word-brushing data corresponding to the sub-speech of the sub-text "At the same time, the manager made a summary and arranged the next plan", so as to facilitate the subsequent simultaneous interpretation quality analysis of each sub-speech.

[0022] Step S13: Analyze and obtain the simultaneous interpretation quality score of the sub-speech based on the brush word data of the sub-speech.

[0023] In the disclosed embodiment, the quality of simultaneous interpretation can be evaluated by several evaluation indicators, and the several evaluation indicators can specifically include: at least one of the word response time, result response time, word brushing ratio, and jump degree, and the evaluation indicators are not limited here. It should be noted that the word response time represents the time interval from the user starting to pronounce (such as the first word or word) to the translation system outputting the translation result for the first time. As a possible example, the word response time can be the primary indicator of the user's perception of the speed of the translation system; the result response time represents the time from the user stopping speaking (such as the last word of a sentence or sentence) to the translation system completing the translation of the segment. As a possible example, the result response time directly affects the completion speed of the translation, especially in continuous dialogues. Too long result response time will lead to interruption of the dialogue rhythm and affect the fluency of communication; the word brushing ratio represents the ratio of the number of output times of the intermediate result (i.e., several texts in the word brushing data) to the number of words (words) of the total text result of the semantic clause when seeing the entire final result of a complete semantic clause; the jump degree represents the average value of the change in the number of words of each jump of the intermediate result (i.e., several texts in the word brushing data) when seeing the entire final result of a complete semantic clause. The specific calculation process of the above evaluation indicators can be referred to the following related descriptions, which will not be repeated here.

[0024] In an implementation scenario, when the simultaneous interpretation quality is evaluated by the word response time, the time when the speaker first utters the word in the sub-speech can be obtained as the first speech time, and the time when the first word is produced in the sub-speech word brushing data can be obtained as the first response time. Then, based on the time difference between the first speech time and the first response time, the simultaneous interpretation quality score of the sub-speech regarding the word response time can be obtained. Please refer to Figure 2b , Figure 2bSchematic diagram of an embodiment of a simultaneous interpretation sequence. Figure 2b As shown, the time when the speaker first utters in sub-speech 1 is t 11 , the time of the first word in the brush word data of sub-speech 1 is t' 11 , the time when the speaker first utters in subspeech 2 is t 21 , the time of the first word in the brush word data of sub-speech 1 is t' 21 , the time when the speaker first utters in sub-speech n-1 is t (n-1)1 , the time of the first word in the brush word data of sub-speech 1 is t' (n-1)1 , the time of the speaker's first utterance in subphone n is t n1 , the time of the first word in the brush word data of sub-speech 1 is t' n1 On this basis, for sub-speech 1, the simultaneous interpretation quality score t' about the word-out response time can be obtained: 11 -t 11 For sub-speech 2, the simultaneous interpretation quality score t' about the word response time can be obtained 21 -t 21 For sub-speech n-1, the simultaneous interpretation quality score t' about the word response time can be obtained (n-1)1 -t (n-1)1 For sub-speech n, the simultaneous interpretation quality score t' about the word response time can be obtained n1 -t n1 . Of course, the above example is only a possible example of the simultaneous interpretation quality score of the first voice time, the first response time and the word response time, and other possible situations will not be given examples one by one here. In the above method, when the simultaneous interpretation quality is evaluated by the word response time, the time when the speaker first utters the word in the sub-voice is obtained as the first voice time, and the time when the first word is produced in the sub-voice word brushing data is obtained as the first response time, and then based on the time difference between the first voice time and the first response time, the simultaneous interpretation quality score of the sub-voice regarding the word response time is obtained, which can evaluate the translation speed of each sub-voice perceived by the user during the simultaneous interpretation process as accurately as possible.

[0025] In an implementation scenario, when the simultaneous interpretation quality is evaluated by the result response time, the time when the speaker last uttered the voice in the sub-voice can be obtained as the second voice time, and the appearance time of the sub-text in the sub-voice brush data can be obtained as the second response time. Then, based on the time difference between the second voice time and the second response time, the simultaneous interpretation quality score of the sub-voice with respect to the result response time can be obtained. Figure 2b , Figure 2b Schematic diagram of an embodiment of a simultaneous interpretation sequence. Figure 2b As shown, the last utterance time of the speaker in sub-speech 1 is t 12, the appearance time of the subtext in the brush word data of sub-voice 1 is t' 12 , the time of the last utterance of the speaker in subspeech 2 is t 22 , the appearance time of the subtext in the brush word data of sub-voice 2 is t' 22 , the time when the speaker last uttered in sub-speech n-1 is t (n-1)2 , the appearance time of the subtext in the brush word data of sub-speech n-1 is t' (n-1)2 , the time of the last utterance of the speaker in subphone n is t n2 , the appearance time of the subtext in the brush word data of subphone n is t' n2 On this basis, for sub-speech 1, the simultaneous interpretation quality score t' about the result response time can be obtained: 12 -t 12 For sub-speech 2, the simultaneous interpretation quality score t' about the result response time can be obtained 22 -t 22 For sub-speech n-1, the simultaneous interpretation quality score t' about the result response time can be obtained (n-1)2 -t (n-1)2 For sub-speech n, the simultaneous interpretation quality score t' about the result response time can be obtained n2 -t n2 . Of course, the above example is only a possible example of the simultaneous interpretation quality score of the second voice time, the second response time and the result response time, and other possible situations will not be given one by one here. In the above method, when the simultaneous interpretation quality is evaluated by the result response time, the time when the speaker's last utterance in the sub-voice is obtained as the second voice time, and the appearance time of the sub-text in the brush word data of the sub-voice is obtained as the second response time, and then based on the time difference between the second voice time and the second response time, the simultaneous interpretation quality score of the sub-voice regarding the result response time is obtained, which can evaluate the completion speed and fluency of the translation of each sub-voice in the simultaneous interpretation process as accurately as possible.

[0026] In an implementation scenario, when the simultaneous interpretation quality is evaluated by the brush word ratio, the total number of texts in the brush word data of the sub-speech can be obtained, and the total length of the sub-texts in the brush word data of the sub-speech can be obtained, and then based on the ratio of the total number to the total length, the simultaneous interpretation quality score of the sub-speech with respect to the brush word ratio can be obtained. Figure 2c , Figure 2c Schematic diagram of another embodiment of brush character data. Figure 2c As shown in the figure, for the sub-text "I am Chinese.", the corresponding sub-speech in the simultaneous interpretation process can contain n texts (i.e. S 1, S 2 … , Sn-1 , S n ). That is, in this example, the total number of several texts in the brush word data of the sub-voice can be recorded as n, and since the sub-text in the brush word data of the sub-voice is "I am Chinese.", its total length can be recorded as On this basis, we can obtain the ratio of the two and get the simultaneous interpretation quality score of the sub-speech with respect to the brush word ratio, that is, . Of course, the above example is only a possible example of the simultaneous interpretation quality score of the brush word ratio, and other possible situations will not be given one by one here. It should be noted that the "brushing word ratio" in the embodiment of the present disclosure can intuitively reflect the jump frequency of the translation result when the simultaneous interpretation function translates a piece of audio. For example, if the brush word ratio is 0.8, it means that on average each word needs to jump 0.8 times to correct it to the final result. That is to say, when 100 words are translated, the intermediate result is output 80 times. In the above method, when the simultaneous interpretation quality is evaluated by the brush word ratio, the total number of several texts in the brush word data of the sub-voice is obtained, and the total length of the sub-text in the brush word data of the sub-voice is obtained, and then based on the ratio of the total number to the total length, the simultaneous interpretation quality score of the sub-voice regarding the brush word ratio is obtained, which can reflect the jump frequency of each sub-voice in the simultaneous interpretation process as accurately as possible.

[0027] In an implementation scenario, when the simultaneous interpretation quality is evaluated by the jump degree, several texts in the brush word data of the sub-voice can be selected as the current text, and the previous text of the current text can be selected as the reference text. Then, based on the total length of the current text and the reference text, the first length is obtained, and the longest common prefix length of the current text and the reference text is obtained as the second length, so that the jump degree of the current text can be obtained based on the first length and the second length, and then the jump degree of each text in the brush word data of the sub-voice can be fused to obtain the simultaneous interpretation quality score of the sub-voice with respect to the jump degree. It should be noted that the jump degree can be used to reflect how many words are changed each time the intermediate result jumps. For example, if the jump degree is 2.6, it means that each time the intermediate result changes, an average of 2.6 words are changed. In the above method, by performing jump analysis on each group of adjacent texts, the simultaneous interpretation quality score of the sub-voice with respect to the jump degree is determined by combining the jump analysis of each group of adjacent texts, and the jump degree of each sub-voice in the simultaneous interpretation process can be analyzed as accurately as possible.

[0028] In a specific implementation scenario, the larger value of the total length of the current text and the reference text can be obtained as the first length, and the jump degree of the current text can be obtained based on the length difference between the first length and the second length. The jump degrees of each text are then averaged to obtain the simultaneous interpretation quality score of the sub-speech regarding the jump degree.

[0029] In a specific implementation scenario, please continue to refer to Figure 2c to select the first text S 1 is taken as the current text as an example. Since the previous text is empty, the larger value of the total length of the current text and the reference text, that is, the first length, is the total length of the current text itself, which can be recorded as L1=Max(l1,0). In addition, since the previous text is empty, the current text and the reference text have no common prefix, so the longest common prefix length is 0, that is, the second length is 0. Therefore, for the first text S 1, its jump degree can be recorded as L1. S 2 Take the current text as an example, then due to its previous text S 1 is l1, so the larger value of the total length of the current text and the reference text, that is, the first length, can be recorded as L2=Max(l2,l1). In addition, the longest common prefix length of the current text and the reference text, that is, the second length, can be recorded as L'2=len(LCP(S2,S1)), where LCP represents the longest common prefix and len(LCP) represents the longest common prefix length. Therefore, for the second text S 2, its jump degree can be recorded as L2-L'2. In the case of selecting other texts as the current text, the same can be applied, and examples are not given here one by one.

[0030] In a specific implementation scenario, Figure 2c Taking the brush word data shown in the figure as an example, the simultaneous interpretation quality score of the sub-speech about the jump degree can be expressed as [L1+(L2-L'2)+(L3-L'3)…+(L n-1 -L' n-1 )+(L n -L' n )] / n. Of course, the above examples are only Figure 2c The brush character data shown is taken as an example, which is a possible example of the jump degree of the sub-speech. Other possible situations will not be given one by one here.

[0031] It should be noted that the above examples are possible examples of obtaining the simultaneous interpretation quality score of the sub-speech when the word response time, result response time, word brushing ratio and jump degree are used as evaluation indicators respectively. The use of other evaluation indicators is not limited thereto. The specific process of obtaining the simultaneous interpretation quality score when other simultaneous interpretation indicators are used will not be given one by one here.

[0032] Step S14: Based on the simultaneous interpretation quality scores of the sub-speech, the target score representing the simultaneous interpretation quality of the target speech is obtained by fusion.

[0033] Specifically, after obtaining the simultaneous interpretation quality scores of each sub-speech, the simultaneous interpretation quality scores of each sub-speech with respect to the same evaluation index can be merged to obtain the simultaneous interpretation quality score of the target speech with respect to the corresponding evaluation index. For example, taking the word response time as an evaluation index, the simultaneous interpretation quality scores of each sub-speech with respect to the word response time can be averaged to obtain the simultaneous interpretation quality score of the target speech with respect to the word response time: =[( - )+( - )+...+( - )+( - )] / n(1) In the above formula (1), Indicates the simultaneous interpretation quality score of the target speech with respect to the word response time. In addition, for the specific meanings of other parameters in formula (1), please refer to the above-mentioned related descriptions, which will not be repeated here. Alternatively, taking the result response time as an evaluation indicator, the simultaneous interpretation quality scores of each sub-speech with respect to the result response time can be averaged to obtain the simultaneous interpretation quality score of the target speech with respect to the result response time: =[( - )+( - )+...+( - )+( - )] / n(2) In the above formula (2), Indicates the simultaneous interpretation quality score of the target speech regarding the result response time. In addition, for the specific meanings of other parameters in formula (2), please refer to the above-mentioned related descriptions, which will not be repeated here. Alternatively, when taking the word-brushing ratio and jump degree as evaluation indicators as an example, the calculation method of the above-mentioned word response time and result response time (such as averaging) can be referred to to obtain the simultaneous interpretation quality score of the target speech regarding the word-brushing ratio and the simultaneous interpretation quality score regarding the jump degree, which will not be repeated here. It should be noted that if the performance in the word response time and the result response time is excellent, and the word-brushing ratio is low and the jump degree is low, it means that the simultaneous interpretation has a fast and stable translation output capability, which can bring users a low-delay and smooth user experience. Please refer to Table 1, which is a schematic table of an embodiment of a reference range of user perception indicators.

[0034] Table 1 Schematic table of an embodiment of a reference range of user perception indicators

[0035] As shown in Table 1, the word response time is within 1 second, which can be considered to be at the optimization level, and within 2 seconds, which can be considered to be at the standard level; the result response time is within 1 second, which can be considered to be at the optimization level, and within 2 seconds, which can be considered to be at the standard level; the jump degree is between 1 and 2, then the output can be considered stable; the brush word ratio is between 0.5 and 1, then the jump frequency can be considered to be moderate. For details, please refer to Table 1, which will not be repeated here. On this basis, the simultaneous interpretation quality score of the target speech with respect to various evaluation indicators can be combined to obtain the target score representing the simultaneous interpretation quality of the target speech. Exemplarily, the simultaneous interpretation quality score can be quantified as a whole according to the weights of various evaluation indicators to obtain the target score representing the simultaneous interpretation quality of the target speech. Specifically, the simultaneous interpretation quality scores of various evaluation indicators can be first unified according to Table 1 or a reference table similar to Table 1 (e.g., unified to within 0 to 10 points), and then the unified simultaneous interpretation quality scores can be weighted by the weights of various evaluation indicators to obtain the target score representing the simultaneous interpretation quality of the target speech.

[0036] In an implementation scenario, please refer to Figure 2d , Figure 2d 1 is a schematic diagram of the process of an embodiment of simultaneous interpretation quality evaluation of the present application. Figure 2d As shown, after the simultaneous interpretation text of the target speech is segmented based on punctuation and semantics, several sub-texts can be obtained. For each sub-text, the brush word data of the corresponding sub-speech in the target speech can be obtained, and the brush word data includes several texts of the sub-speech in the simultaneous interpretation process from the first word to the gradual correction until the final translation into the sub-text. Then, based on the brush word data of the sub-speech, the simultaneous interpretation quality score of the sub-speech is analyzed. On this basis, the simultaneous interpretation quality scores of each sub-speech can be integrated to obtain the target score that represents the simultaneous interpretation quality of the target speech. It should be noted that although Figure 2d In the brush word data of each sub-voice, several texts are represented as "text 1, text 2, ..., text m", but this does not limit the number of texts in the brush word data of each sub-voice to be the same. In actual application, the number of texts in the brush word data of each sub-voice can be completely different; or, the number of texts in the brush word data of each sub-voice can be not completely the same, that is, the brush word data of some sub-voices have the same number of texts, while the brush word data of some sub-voices have different numbers of texts. The number of texts in the brush word data of each sub-voice is not limited here.

[0037] The above scheme is based on segmenting the simultaneous interpretation text of the target speech to obtain several sub-texts, and obtaining the brush word data of the sub-speech corresponding to the sub-text in the target speech, and the brush word data of the sub-speech includes: several texts of the sub-speech from the first word to the gradual correction until the final translation into the sub-text during the simultaneous interpretation process, and then based on the brush word data of the sub-speech, the simultaneous interpretation quality score of the sub-speech is analyzed, and then based on the simultaneous interpretation quality scores of each sub-speech, the target score representing the simultaneous interpretation quality of the target speech is obtained by fusion. Therefore, on the one hand, by dividing the simultaneous interpretation text into several sub-texts, and then performing simultaneous interpretation on each sub-text corresponding to the sub-speech Compared with the overall quality evaluation of the target speech and its simultaneous interpretation text, the simultaneous interpretation quality evaluation and final score fusion can further refine the evaluation granularity, which helps to improve the accuracy of the simultaneous interpretation quality evaluation to a certain extent. On the other hand, when evaluating each sub-speech, the brush word data of the sub-speech is combined, and the brush word data includes the sub-speech from the first word to the gradual correction and finally the translation into a sub-text during the simultaneous interpretation process. Therefore, in the process of simultaneous interpretation quality evaluation of the sub-speech, attention can be paid to the simultaneous interpretation process, which helps to measure the simultaneous interpretation quality in the brush word process in the application scenario of streaming simultaneous interpretation. Therefore, the accuracy of simultaneous interpretation quality evaluation can be improved in the application scenario of streaming simultaneous interpretation.

[0038] See also Figure 3 , Figure 3 It is a schematic diagram of the framework of an embodiment of the simultaneous interpretation quality evaluation device of the present application. The simultaneous interpretation quality evaluation device 30 includes: a text segmentation module 31, a data acquisition module 32, a quality analysis module 33 and a score fusion module 34. The text segmentation module 31 is used to segment the simultaneous interpretation text based on the target speech to obtain a plurality of sub-texts; the data acquisition module 32 is used to obtain the brush word data of the sub-speech corresponding to the sub-text in the target speech; wherein the brush word data of the sub-speech includes: a plurality of texts of the sub-speech from the first word output to the gradual correction until the final translation into the sub-text during the simultaneous interpretation process; the quality analysis module 33 is used to analyze the brush word data of the sub-speech to obtain the simultaneous interpretation quality score of the sub-speech based on the sub-speech; the score fusion module 34 is used to fuse the simultaneous interpretation quality scores of each sub-speech to obtain the target score representing the simultaneous interpretation quality of the target speech based on the simultaneous interpretation quality scores of each sub-speech.

[0039] In the above scheme, the simultaneous interpretation quality evaluation device 30 divides the simultaneous interpretation text of the target speech into several sub-texts, obtains the brush word data of the sub-speech corresponding to the sub-text in the target speech, and the brush word data of the sub-speech includes: several texts of the sub-speech from the first word to the gradual correction until the final translation into the sub-text in the simultaneous interpretation process, and then analyzes and obtains the simultaneous interpretation quality score of the sub-speech based on the brush word data of the sub-speech, and then integrates the simultaneous interpretation quality scores of each sub-speech to obtain the target score representing the simultaneous interpretation quality of the target speech. Therefore, on the one hand, by dividing the simultaneous interpretation text into several sub-texts, and then respectively corresponding to the sub-texts for each sub-text, Compared with the overall quality evaluation of the target speech and its simultaneous interpretation text, the simultaneous interpretation quality evaluation of the speech and the final score fusion can further refine the evaluation granularity, which helps to improve the accuracy of the simultaneous interpretation quality evaluation to a certain extent. On the other hand, when evaluating each sub-speech, the brush word data of the sub-speech is combined, and the brush word data includes the sub-speech from the first word to the gradual correction and finally the translation into a sub-text during the simultaneous interpretation process. Therefore, in the process of evaluating the simultaneous interpretation quality of the sub-speech, the simultaneous interpretation process can be paid attention to, which helps to measure the simultaneous interpretation quality in the brush word process in the application scenario of streaming simultaneous interpretation. Therefore, the accuracy of the simultaneous interpretation quality evaluation can be improved in the application scenario of streaming simultaneous interpretation.

[0040] In some disclosed embodiments, the text segmentation module 31 includes a first segmentation submodule, which is used to segment the simultaneous interpretation text based on punctuation to obtain a number of first sub-texts; the text segmentation module 31 includes a second segmentation submodule, which is used to segment the simultaneous interpretation text based on semantics to obtain a number of second sub-texts; the text segmentation module 31 includes a segmentation integration submodule, which is used to integrate a number of first sub-texts and a number of second sub-texts to obtain a number of sub-texts.

[0041] In some disclosed embodiments, the quality analysis module 33 includes a word-out response evaluation submodule, which is used to obtain the time when the speaker first utters the word in the sub-speech as the first speech time, and obtain the time when the first word is produced in the word-out data of the sub-speech as the first response time when the simultaneous interpretation quality is evaluated by the word-out response time; the quality analysis module 33 includes a word-out quality scoring submodule, which is used to obtain the simultaneous interpretation quality score of the sub-speech regarding the word-out response time based on the time difference between the first speech time and the first response time.

[0042] In some disclosed embodiments, the quality analysis module 33 includes a result response evaluation submodule, which is used to obtain the time when the speaker's last utterance in the sub-speech is used as the second speech time, and obtain the appearance time of the sub-text in the brush word data of the sub-speech as the second response time when the simultaneous interpretation quality is evaluated by the result response time; the quality analysis module 33 includes a result quality scoring submodule, which is used to obtain the simultaneous interpretation quality score of the sub-speech with respect to the result response time based on the time difference between the second speech time and the second response time.

[0043] In some disclosed embodiments, the quality analysis module 33 includes a text information acquisition submodule, which is used to obtain the total number of several texts in the brush word data of the sub-speech and the total length of the sub-texts in the brush word data of the sub-speech when the simultaneous interpretation quality is evaluated by the brush word ratio; the quality analysis module 33 includes a brush word quality scoring submodule, which is used to obtain the simultaneous interpretation quality score of the sub-speech regarding the brush word ratio based on the ratio of the total number to the total length.

[0044] In some disclosed embodiments, the quality analysis module 33 includes a text selection submodule, which is used to select several texts in the brush word data of the sub-voice as the current text, and select the previous text of the current text as the reference text when the simultaneous interpretation quality is evaluated by the jump degree; the quality analysis module 33 includes a length measurement submodule, which is used to obtain a first length based on the total length of the current text and the reference text, and obtain the longest common prefix length of the current text and the reference text as the second length; the quality analysis module 33 includes a jump measurement submodule, which is used to obtain the jump degree of the current text based on the first length and the second length; the quality analysis module 33 includes a jump scoring submodule, which is used to fuse the jump degrees of each text in the brush word data of the sub-voice based on the jump degree, and obtain the simultaneous interpretation quality score of the sub-voice with respect to the jump degree.

[0045] In some disclosed embodiments, the length measurement submodule is specifically used to obtain the larger value of the total length of the current text and the reference text as the first length; the jump measurement submodule is specifically used to obtain the jump degree of the current text based on the length difference between the first length and the second length; the jump scoring submodule is specifically used to average the jump degrees of each text to obtain the simultaneous interpretation quality score of the sub-speech regarding the jump degree.

[0046] In some disclosed embodiments, the scoring fusion module 34 includes a scoring fusion submodule, which is used to fuse the simultaneous interpretation quality scores of each sub-speech with respect to the same evaluation index to obtain the simultaneous interpretation quality score of the target speech with respect to the corresponding evaluation index; the scoring fusion module 34 includes a scoring synthesis submodule, which is used to synthesize the simultaneous interpretation quality scores of the target speech with respect to various evaluation indicators to obtain a target score representing the simultaneous interpretation quality of the target speech.

[0047] In some disclosed embodiments, the quality of simultaneous interpretation is evaluated using a number of evaluation indicators, and the number of evaluation indicators include: at least one of word response time, result response time, word brushing ratio, and jump degree; and / or, the simultaneous interpretation text is obtained by simultaneously interpreting the target speech using a large simultaneous interpretation model.

[0048] See also Figure 4 , Figure 4 4 is a schematic diagram of a framework of an embodiment of an electronic device of the present application. The electronic device 40 includes at least a memory 41 and a processor 42 coupled to each other, the memory 41 stores at least program instructions, and the processor 42 is used to execute the program instructions to implement the steps in any of the above-mentioned simultaneous interpretation quality evaluation method embodiments. For details, please refer to the aforementioned disclosed embodiments, which will not be repeated here. As a possible example, the electronic device 40 may include but is not limited to a server, etc., and the specific type of the electronic device 40 is not limited here.

[0049] Specifically, the processor 42 is used to control itself and the memory 41 to implement the steps in any of the above-mentioned simultaneous interpretation quality evaluation method embodiments. The processor 42 can also be called a CPU (Central Processing Unit). The processor 42 may be an integrated circuit chip with signal processing capabilities. The processor 42 can also be a general-purpose processor, a digital signal processor (Digital Signal Processor, DSP), an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field-programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. In addition, the processor 42 can be implemented by an integrated circuit chip.

[0050] In the above scheme, the electronic device 40 divides the simultaneous interpretation text of the target speech to obtain a plurality of sub-texts, obtains the brush word data of the sub-speech corresponding to the sub-text in the target speech, and the brush word data of the sub-speech includes: a plurality of texts of the sub-speech from the first word output to the gradual correction until the final translation into the sub-text in the simultaneous interpretation process, and then analyzes and obtains the simultaneous interpretation quality score of the sub-speech based on the brush word data of the sub-speech, and then integrates and obtains the target score representing the simultaneous interpretation quality of the target speech based on the simultaneous interpretation quality scores of each sub-speech, so on the one hand, by dividing the simultaneous interpretation text into a plurality of sub-texts, and then corresponding to the sub-speech for each sub-text, Compared with the overall quality evaluation of the target speech and its simultaneous interpretation text, the simultaneous interpretation quality evaluation and final score fusion can further refine the evaluation granularity, which helps to improve the accuracy of the simultaneous interpretation quality evaluation to a certain extent. On the other hand, when evaluating each sub-speech, the sub-speech word brushing data is combined, and the word brushing data includes the sub-speech from the first word to the gradual correction and finally the translation into a sub-text during the simultaneous interpretation process. Therefore, in the process of simultaneous interpretation quality evaluation of the sub-speech, attention can be paid to the simultaneous interpretation process, which helps to measure the simultaneous interpretation quality in the word brushing process in the application scenario of streaming simultaneous interpretation. Therefore, the accuracy of simultaneous interpretation quality evaluation can be improved in the application scenario of streaming simultaneous interpretation.

[0051] See also Figure 5 , Figure 5 1 is a schematic diagram of a framework of an embodiment of a computer-readable storage medium 50 of the present application. The computer-readable storage medium 50 stores program instructions 51 that can be executed by a processor, and the program instructions 51 are used to implement the steps in any of the above-mentioned simultaneous interpretation quality evaluation method embodiments.

[0052] In the above scheme, the computer-readable storage medium 50 is segmented based on the simultaneous interpretation text of the target speech to obtain a plurality of subtexts, and the brush word data of the sub-speech corresponding to the subtext in the target speech is obtained, and the brush word data of the sub-speech includes: a plurality of texts of the sub-speech from the first word to the gradual correction until the final translation into the subtext during the simultaneous interpretation process, and then based on the brush word data of the sub-speech, the simultaneous interpretation quality score of the sub-speech is analyzed, and then based on the simultaneous interpretation quality scores of each sub-speech, the target score representing the simultaneous interpretation quality of the target speech is obtained by fusion. Therefore, on the one hand, by dividing the simultaneous interpretation text into a plurality of subtexts, and then corresponding to each subtext, the brush word data of the sub-speech includes: a plurality of texts of the sub-speech from the first word to the gradual correction until the final translation into the subtext, and then based on the brush word data of the sub-speech, the simultaneous interpretation quality score of the sub-speech is obtained by fusion. Compared with the overall quality evaluation of the target speech and its simultaneous interpretation text, the simultaneous interpretation quality evaluation of the speech and the final score fusion can further refine the evaluation granularity, which helps to improve the accuracy of the simultaneous interpretation quality evaluation to a certain extent. On the other hand, when evaluating each sub-speech, the brush word data of the sub-speech is combined, and the brush word data includes the sub-speech from the first word to the gradual correction and finally the translation into a sub-text during the simultaneous interpretation process. Therefore, in the process of evaluating the simultaneous interpretation quality of the sub-speech, the simultaneous interpretation process can be paid attention to, which helps to measure the simultaneous interpretation quality in the brush word process in the application scenario of streaming simultaneous interpretation. Therefore, the accuracy of the simultaneous interpretation quality evaluation can be improved in the application scenario of streaming simultaneous interpretation.

[0053] In some embodiments, the functions or modules included in the device provided by the embodiments of the present disclosure can be used to execute the method described in the above method embodiments. The specific implementation can refer to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0054] The above description of various embodiments tends to emphasize the differences between the various embodiments. The same or similar aspects can be referenced to each other, and for the sake of brevity, they will not be repeated herein.

[0055] In the several embodiments provided in the present application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device implementation described above is only schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0056] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0057] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0058] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of each implementation method of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program code.

[0059] If the technical solution of this application involves personal information, the product using the technical solution of this application has clearly informed the personal information processing rules and obtained the individual's voluntary consent before processing the personal information. If the technical solution of this application involves sensitive personal information, the product using the technical solution of this application has obtained the individual's separate consent before processing the sensitive personal information, and at the same time meets the "explicit consent" requirement. For example, on personal information collection devices such as cameras, clear and prominent signs are set to inform that the personal information collection scope has been entered and personal information will be collected. If the individual voluntarily enters the collection scope, it is deemed that he or she agrees to the collection of his or her personal information; or on the device that processes personal information, the personal information processing rules are notified by obvious signs / information, and the individual's authorization is obtained through pop-up information or by asking the individual to upload his or her personal information; among them, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the type of personal information processed.

Claims

1. A method for evaluating simultaneous interpretation quality, characterized in that: include: The simultaneous interpretation text based on the target speech is segmented to obtain several sub-texts; Acquire the brush word data of the sub-speech corresponding to the sub-text in the target speech; wherein the brush word data of the sub-speech includes: a plurality of texts of the sub-speech from the first word generation to the gradual correction and finally translation into the sub-text in the simultaneous interpretation process; Based on the brush word data of the sub-speech, analyzing and obtaining the simultaneous interpretation quality score of the sub-speech; Based on the simultaneous interpretation quality scores of the sub-speech, a target score representing the simultaneous interpretation quality of the target speech is obtained by fusion.

2. The method according to claim 1, characterized in that The simultaneous interpretation text based on the target speech is segmented to obtain several sub-texts, including: Segmenting the simultaneous interpretation text based on punctuation to obtain a plurality of first sub-texts, and segmenting the simultaneous interpretation text based on semantics to obtain a plurality of second sub-texts; The plurality of subtexts are obtained by integrating the plurality of first subtexts and the plurality of second subtexts.

3. The method according to claim 1, characterized in that In the case where the simultaneous interpretation quality is evaluated by the word output response time, the simultaneous interpretation quality score of the sub-speech is obtained by analyzing the word output data based on the sub-speech, including: The time when the speaker first utters a sound in the sub-speech is obtained as the first speech time, and the time when the first word is produced in the brush word data of the sub-speech is obtained as the first response time; Based on the time difference between the first speech time and the first response time, a simultaneous interpretation quality score of the sub-speech with respect to the word output response time is obtained.

4. The method according to claim 1, characterized in that: In the case where the simultaneous interpretation quality is evaluated by the result response time, the simultaneous interpretation quality score of the sub-speech is obtained by analyzing the brush word data based on the sub-speech, including: Acquire the last utterance time of the speaker in the sub-speech as the second speech time, and acquire the appearance time of the sub-text in the brush word data of the sub-speech as the second response time; Based on the time difference between the second speech time and the second response time, a simultaneous interpretation quality score of the sub-speech with respect to the result response time is obtained.

5. The method according to claim 1, characterized in that When the simultaneous interpretation quality is evaluated by the word-brushing ratio, the simultaneous interpretation quality score of the sub-speech is obtained by analyzing the word-brushing data of the sub-speech, including: Obtaining the total number of the plurality of texts in the brush word data of the sub-voice, and obtaining the total length of the sub-texts in the brush word data of the sub-voice; Based on the ratio of the total number to the total length, a simultaneous interpretation quality score of the sub-speech with respect to the brush-word ratio is obtained.

6. The method according to claim 1, characterized in that In the case where the simultaneous interpretation quality is evaluated by the jump degree, the simultaneous interpretation quality score of the sub-speech is obtained by analyzing the brush word data based on the sub-speech, including: Selecting the plurality of texts in the brush word data of the sub-voice respectively as current texts, and selecting the text preceding the current text as reference text; Based on the total length of each of the current text and the reference text, a first length is obtained, and the longest common prefix length of the current text and the reference text is obtained as a second length; Based on the first length and the second length, obtaining the jump degree of the current text; The jump degree of each of the texts in the brush word data of the sub-voice is fused to obtain the simultaneous interpretation quality score of the sub-voice with respect to the jump degree.

7. The method according to claim 6, characterized in that The obtaining a first length based on the total length of each of the current text and the reference text comprises: Obtaining a larger value of the total length of the current text and the reference text as the first length; And / or, obtaining the jump degree of the current text based on the first length and the second length includes: Based on the length difference between the first length and the second length, obtaining the jump degree of the current text; And / or, the transition degrees of each of the texts in the brush character data based on the sub-voice are merged to obtain the simultaneous interpretation quality score of the sub-voice with respect to the transition degree, including: Based on the average of the jump degrees of each of the texts, a simultaneous interpretation quality score of the sub-speech with respect to the jump degree is obtained.

8. The method according to claim 1, characterized in that The simultaneous interpretation quality scores based on the sub-speech are integrated to obtain a target score representing the simultaneous interpretation quality of the target speech, including: Based on the simultaneous interpretation quality scores of the sub-speech with respect to the same evaluation index, the simultaneous interpretation quality score of the target speech with respect to the corresponding evaluation index is obtained by fusing; The simultaneous interpretation quality scores of the target speech with respect to various evaluation indicators are integrated to obtain a target score representing the simultaneous interpretation quality of the target speech.

9. The method according to any one of claims 1 to 8, characterized in that: The simultaneous interpretation quality is evaluated by a plurality of evaluation indicators, and the plurality of evaluation indicators include: at least one of a word output response time, a result response time, a word brushing ratio, and a jump degree; And / or, the simultaneous interpretation text is obtained by simultaneously interpreting the target speech by a large simultaneous interpretation model.

10. A simultaneous interpretation quality evaluation device, characterized in that: include: The text segmentation module is used to segment the simultaneous interpretation text based on the target speech to obtain several sub-texts; A data acquisition module is used to acquire the brush word data of the sub-speech corresponding to the sub-text in the target speech; wherein the brush word data of the sub-speech includes: a plurality of texts of the sub-speech from the first word generation to the gradual correction and finally translation into the sub-text during the simultaneous interpretation process; A quality analysis module, used for analyzing the word-brushing data of the sub-speech to obtain a simultaneous interpretation quality score of the sub-speech; The scoring fusion module is used to obtain a target score representing the simultaneous interpretation quality of the target speech based on the simultaneous interpretation quality scores of the sub-speech.

11. An electronic device, characterized in that: The method comprises at least a memory and a processor coupled to each other, wherein the memory stores at least program instructions, and the processor is used to execute the program instructions to implement the simultaneous interpretation quality evaluation method according to any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that: Program instructions that can be executed by a processor are stored, and the program instructions are used to implement the simultaneous interpretation quality evaluation method according to any one of claims 1 to 9.

Images