Speech recognition result detection method and apparatus, and storage medium

Abstract

A speech recognition result detection method and device and a storage medium, the method comprising: obtaining a to-be-tested speech recognition result, and translating the to-be-tested speech recognition result from a first language to a second language using a machine translation model to obtain a to-be-tested translation result (S101); determining a first evaluation feature based on the to-be-tested translation result and the to-be-tested speech recognition result, the first evaluation feature being used to represent the recognition and translation effects of the to-be-tested speech recognition result (S102); evaluating the to-be-tested speech recognition result based on the first evaluation feature to obtain a first evaluation result (S103); determining whether the first evaluation result satisfies a preset condition to obtain a first judgment result, and determining an error detection result of the to-be-tested speech recognition result according to the first judgment result (S104).

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the technical field of speech recognition, and in particular, to a speech recognition result detection method and device, and a storage medium. BACKGROUND

[0002] Due to the accuracy of speech recognition, errors often occur in speech recognition results, which will hinder subsequent understanding and increase the difficulty of speech understanding. Generally, after recognizing speech and obtaining a speech recognition result, error detection is performed on the speech recognition result, so that errors therein can be further corrected.

[0003] Currently, a corresponding error detection model is trained using some language features of the language to which the speech recognition result belongs, to perform error detection on the speech recognition result. For example, for a Chinese speech recognition result, an error detection model can be designed based on word collocation and context information to implement error detection of the Chinese speech recognition result. However, using an error detection model corresponding to the language to which the speech recognition result belongs for error detection can use fewer detection features and has lower detection accuracy. SUMMARY

[0004] To solve the problems in the related art, embodiments of the present application provide a speech recognition result detection method and device, and a storage medium.

[0005] The technical solution of the embodiments of the present application can be implemented as follows:

[0006] The embodiments of the present application provide a speech recognition result detection method, comprising:

[0007] obtaining a to-be-tested speech recognition result, and translating the to-be-tested speech recognition result from a first language to a second language using a machine translation model to obtain a to-be-tested translation result;

[0008] determining a first evaluation feature based on the to-be-tested translation result and the to-be-tested speech recognition result; the first evaluation feature is used to represent the recognition and translation effects of the to-be-tested speech recognition result;

[0009] evaluating the to-be-tested speech recognition result based on the first evaluation feature to obtain a first evaluation result;

[0010] determining whether the first evaluation result meets a preset condition to obtain a first judgment result, and determining an error detection result of the to-be-tested speech recognition result according to the first judgment result.

[0011] In the above solution, the first evaluation feature is determined based on the to-be-tested translation result and the to-be-tested speech recognition result, comprising:

[0012] obtain a first translation score and a first perplexity score by acquiring a translation score and a perplexity score of the machine translation model for the to-be-tested translation result;

[0013] input the to-be-tested speech recognition result into a language model corresponding to the first language to obtain a first language model score;

[0014] determine the first translation score, the first perplexity score and the first language model score as the first evaluation feature.

[0015] In the above solution, the evaluating the to-be-tested speech recognition result based on the first evaluation feature to obtain a first evaluation result comprises:

[0016] obtain three feature weights by acquiring a feature weight corresponding to each of the first translation score, the first perplexity score and the first language model score;

[0017] obtain the first evaluation result by performing weighted processing on the first translation score, the first perplexity score and the first language model score using the three feature weights.

[0018] In the above solution, before the obtaining three feature weights by acquiring a feature weight corresponding to each of the first translation score, the first perplexity score and the first language model score, the method further comprises:

[0019] obtain a sample speech recognition result and a preset detection result of the sample speech recognition result;

[0020] obtain a sample translation result by translating the sample speech recognition result from the first language to the second language using the machine translation model;

[0021] determine a second translation score, a second perplexity score and a second language model score based on the sample translation result and the sample speech recognition result;

[0022] obtain three preset weights by acquiring a preset weight corresponding to each of the second translation score, the second perplexity score and the second language model score;

[0023] determine the three feature weights using the second translation score, the second perplexity score, the second language model score, the preset detection result and the three preset weights.

[0024] In the above solution, the determining the three feature weights using the second translation score, the second perplexity score, the second language model score, the preset detection result and the three preset weights comprises:

[0025] The second translation score, the second perplexity score, and the second language model score are weighted using the three preset weights to obtain a second evaluation result; the second evaluation feature is used to characterize the recognition and translation effect of the sample speech recognition result;

[0026] Determining whether the second evaluation result meets the preset condition to obtain a second determination result, and determining an error detection result of the sample speech recognition result according to the second determination result;

[0027] Based on the error detection result of the sample speech recognition result and the preset detection result, the three preset weights are adjusted to obtain the three feature weights.

[0028] In the above solution, the three preset weights are adjusted based on the error detection result of the sample speech recognition result and the preset detection result to obtain the three feature weights, including:

[0029] The three preset weights are adjusted according to a weight adjustment algorithm until the error detection result of the sample speech recognition result is the same as the preset detection result, thereby obtaining the three feature weights.

[0030] In the above solution, determining the error detection result of the speech recognition result to be tested according to the first judgment result includes:

[0031] If the first judgment result is that the first evaluation result does not meet the preset condition, determining that the error detection result of the speech recognition result to be tested is an error;

[0032] In a case where the first judgment result is that the first evaluation result satisfies the preset condition, it is determined that the error detection result of the speech recognition result to be tested is error-free.

[0033] In the above solution, determining whether the first evaluation result meets a preset condition to obtain a first determination result includes:

[0034] comparing the first evaluation result with an evaluation threshold;

[0035] If the first evaluation result is less than the evaluation threshold, determining the first judgment result is that the first evaluation result does not meet the preset condition;

[0036] In a case where the first evaluation result is greater than or equal to the evaluation threshold, the first judgment result is determined to be that the first evaluation result meets the preset condition.

[0037] The present invention provides a speech recognition result detection device, comprising:

[0038] a translation module configured to obtain a to-be-tested speech recognition result, and translate the to-be-tested speech recognition result from a first language to a second language using a machine translation model to obtain a to-be-tested translation result;

[0039] a determination module configured to determine a first evaluation feature based on the to-be-tested translation result and the to-be-tested speech recognition result, the first evaluation feature being used to represent recognition and translation effects of the to-be-tested speech recognition result;

[0040] an evaluation module configured to evaluate the to-be-tested speech recognition result based on the first evaluation feature to obtain a first evaluation result;

[0041] a judgment module configured to judge whether the first evaluation result meets a preset condition to obtain a first judgment result, and determine an error detection result of the to-be-tested speech recognition result according to the first judgment result.

[0042] Embodiments of the present application provide a speech recognition result detection device, the device comprising a processor and a memory;

[0043] The processor is configured to execute a speech recognition result detection program stored in the memory to implement the speech recognition result detection method described above.

[0044] Embodiments of the present application provide a computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the speech recognition result detection method described above.

[0045] Embodiments of the present application provide a speech recognition result detection method and device, and a storage medium. The method comprises: obtaining a to-be-tested speech recognition result, and translating the to-be-tested speech recognition result from a first language to a second language using a machine translation model to obtain a to-be-tested translation result; determining a first evaluation feature based on the to-be-tested translation result and the to-be-tested speech recognition result, the first evaluation feature being used to represent recognition and translation effects of the to-be-tested speech recognition result; evaluating the to-be-tested speech recognition result based on the first evaluation feature to obtain a first evaluation result; judging whether the first evaluation result meets a preset condition to obtain a first judgment result, and determining an error detection result of the to-be-tested speech recognition result according to the first judgment result. The technical solution provided in the embodiments of the present application translates the to-be-tested speech recognition result into another language to perform error detection on the to-be-tested speech recognition result in combination with the translation result. Compared with error detection based on only a single language-related feature, more features can be obtained to implement error detection, and the accuracy of error detection is improved. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 A flowchart of a speech recognition result detection method provided in embodiments of the present application is shown.

[0047] Figure 2 An exemplary voice recognition result error detection process schematic diagram provided for an embodiment of the present application;

[0048] Figure 3 A flowchart of determining three feature weights provided for an embodiment of the present application;

[0049] Figure 4 An exemplary voice recognition result detection method application system architecture schematic diagram provided for an embodiment of the present application;

[0050] Figure 5 A structure schematic of a voice recognition result detection device provided for an embodiment of the present application Figure 1 ;

[0051] Figure 6 A structure schematic of a voice recognition result detection device provided for an embodiment of the present application Figure 2 . DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. It can be understood that the specific embodiments described herein are only used to explain the related application, but not to limit the application. In addition, it should be noted that, for the convenience of description, only the parts related to the application are shown in the drawings.

[0053] The present application provides a voice recognition result detection method, which is implemented by a voice recognition result detection device. Figure 1 A flowchart of a voice recognition result detection method provided for an embodiment of the present application. As shown in the figure, the voice recognition result detection method mainly includes the following steps: Figure 1

[0054] S101, obtain a to-be-tested voice recognition result, and use a machine translation model to translate the to-be-tested voice recognition result from a first language to a second language to obtain a to-be-tested translation result.

[0055] In the embodiments of the present application, the voice recognition result detection device can obtain a to-be-tested voice recognition result, and use a machine translation model to translate the to-be-tested voice recognition result from a first language to a second language to obtain a to-be-tested translation result.

[0056] ​It should be noted that in the embodiments of the present application, the to-be-tested speech recognition result is a speech recognition result that needs to be error detected, and is a text generated after speech recognition. The speech recognition result detection apparatus can receive the to-be-tested speech recognition result obtained by the speech recognition processing apparatus, and in addition, the speech recognition result detection apparatus itself can also be configured with an identification module to perform speech recognition to obtain the to-be-tested speech recognition result. The specific to-be-tested speech recognition result and the source of the to-be-tested speech recognition result are not limited in the embodiments of the present application.

[0057] It should be noted that in the embodiments of the present application, the language of the to-be-tested speech recognition result is a first language. The machine translation model is a model that can translate input text from the first language to a second language, and the first language and the second language are two different languages. The specific machine translation model and the first language and the second language can be selected according to actual needs, and the embodiments of the present application are not limited.

[0058] For example, in the embodiments of the present application, the language of the to-be-tested speech recognition result is Chinese, that is, the first language is Chinese, and the machine translation model can translate the to-be-tested speech recognition result from Chinese to English, that is, the second language is English, so as to obtain the to-be-tested translation result in English.

[0059] S102, based on the to-be-tested translation result and the to-be-tested speech recognition result, determining a first evaluation feature; the first evaluation feature is used to represent the recognition and translation effect of the to-be-tested speech recognition result.

[0060] In the embodiments of the present application, after the speech recognition result detection apparatus obtains the to-be-tested speech recognition result and the to-be-tested speech recognition result, the speech recognition result detection apparatus can determine a first evaluation feature based on the to-be-tested translation result and the to-be-tested speech recognition result, and the first evaluation feature is used to represent the recognition and translation effect of the to-be-tested speech recognition result.

[0061] Specifically, in the embodiments of the present application, the speech recognition result detection apparatus determines a first evaluation feature based on the to-be-tested translation result and the to-be-tested speech recognition result, including: obtaining a first translation score and a first perplexity score of the to-be-tested translation result by the machine translation model; inputting the to-be-tested speech recognition result into a language model corresponding to the first language to obtain a first language model score; and determining the first translation score, the first perplexity score and the first language model score as the first evaluation feature.

[0062] It should be noted that in the embodiments of the present application, in the process of the speech recognition result detection apparatus translating the to-be-tested speech recognition result by using the machine translation model in step S101, actually, a plurality of translation results in the second language can be translated, and the corresponding translation score and perplexity score can be determined for each translation result. The translation score represents the overall translation effect of the translation result, for example, the fluency degree and the semantic matching degree, and the perplexity score represents the confusion degree of the translation result in the translation process. The machine translation model actually determines the translation result with the highest translation score in the plurality of translation results as the to-be-tested translation result corresponding to the to-be-tested speech recognition result, and thus the speech recognition result detection apparatus can directly obtain the translation score and the perplexity score of the to-be-tested translation result by the machine translation model, and determine the first translation score and the first perplexity score.

[0063] It should be noted that in the embodiments of the present application, the language of the to-be-tested speech recognition result is the first language, and thus the speech recognition result detection apparatus can input the to-be-tested speech recognition result into the language model corresponding to the first language. The language model can evaluate the to-be-tested speech recognition result from the fluency, the sentence structure and the like based on the designed word collocation and context structure related to the first language, so as to obtain the corresponding language model score. That is, the first speech model score represents the semantic expression effect of the to-be-tested speech recognition result in the first language environment.

[0064] Exemplarily, in the embodiments of the present application, the to-be-tested speech recognition result is “today how is the weather”, and the speech recognition result detection apparatus can input the to-be-tested speech recognition result into the language model of the Chinese language. The model evaluates from the sentence fluency and structure, so as to obtain the first language model score A1. In addition, the speech recognition result detection apparatus translates “today how is the weather” to English by using the machine translation model, obtains the to-be-tested translation result as “What’s the weather like today”, and the first translation score A2 and the first perplexity score A3 of the to-be-tested translation result. A1, A2 and A3 are all the first evaluation features.

[0065] It can be understood that in the embodiments of the present application, the speech recognition result detection apparatus determines the first translation score, the first perplexity score and the first language model score as the first evaluation features, which not only considers the recognition effect from the language characteristics of the to-be-tested speech recognition result, but also considers the expression result of another language of the to-be-tested speech recognition result, so as to obtain more information representing the recognition effect of the to-be-tested speech recognition result, and thus the first evaluation features can be used to more accurately perform the subsequent error detection.

[0066] It should be noted that in the embodiments of the present application, the speech recognition result detection apparatus can also determine the first evaluation features based on the to-be-tested translation result and the to-be-tested speech recognition result in other manners and from other perspectives. For example, the speech recognition result detection apparatus can also score the to-be-tested translation result and the to-be-tested speech recognition result from the aspects of fluency, language logic, etc. by using corresponding models, and determine each score as a first evaluation feature. The number and type of specific first evaluation features are not limited in the embodiments of the present application.

[0067] S103, evaluating the to-be-tested speech recognition result based on the first evaluation features to obtain a first evaluation result.

[0068] In the embodiments of the present application, after obtaining the first evaluation features, the speech recognition result detection apparatus further evaluates the to-be-tested speech recognition result based on the first evaluation features to obtain a first evaluation result.

[0069] Specifically, in the embodiments of the present application, the speech recognition result detection apparatus evaluates the to-be-tested speech recognition result based on the first evaluation features to obtain a first evaluation result, including: obtaining a feature weight corresponding to each of the first translation score, the first perplexity score and the first language model score to obtain three feature weights; and performing weighted processing on the first translation score, the first perplexity score and the first language model score by using the three feature weights to obtain the first evaluation result.

[0070] It should be noted that in the embodiments of the present application, for each type of evaluation feature in the first evaluation features, a corresponding feature weight is set to represent the importance of using the evaluation feature in the process of evaluating the to-be-tested speech recognition result. The speech recognition result detection apparatus can multiply each evaluation feature in the first evaluation features with the corresponding feature weight to obtain three product results, and then calculate the sum of the three product results as the first evaluation result.

[0071] Specifically, in the embodiments of the present application, the first evaluation features include a translation score, a perplexity score and a language model score, and the speech recognition result detection apparatus calculates the first evaluation result of the to-be-tested speech recognition result according to the following formula (1):

[0072] Score(s)=λ1×TM+λ2×PP+λ3×LM(1)

[0073] Wherein, Score(s) is the first evaluation result, TM is the translation score, λ1 is the feature weight corresponding to the translation score, PP is the perplexity score, λ2 is the feature weight corresponding to the perplexity score, LM is the speech model score, and λ3 is the feature weight corresponding to the language model score.

[0074] It should be noted that in the embodiments of the present application, the speech recognition result detection apparatus performs weighting processing on the first evaluation features, and not only can the first evaluation result be obtained in the manner of the above-mentioned weighted summation, but also other forms of weighting processing, such as weighted averaging, can be performed, and the embodiments of the present application are not limited in this regard.

[0075] In S104, it is judged whether the first evaluation result satisfies a preset condition, a first judgment result is obtained, and an error detection result of the to-be-tested speech recognition result is determined according to the first judgment result.

[0076] In the embodiments of the present application, after the first evaluation result is obtained, the speech recognition result detection apparatus can judge whether the first evaluation result satisfies a preset condition, obtain a first judgment result, and determine an error detection result of the to-be-tested speech recognition result according to the first judgment result.

[0077] Specifically, in the embodiments of the present application, the speech recognition result detection apparatus determines the error detection result of the to-be-tested speech recognition result according to the first judgment result, including: in a case where the first judgment result is that the first evaluation result does not satisfy the preset condition, determining that the error detection result of the to-be-tested speech recognition result is that there is an error; and in a case where the first judgment result is that the first evaluation result satisfies the preset condition, determining that the error detection result of the to-be-tested speech recognition result is that there is no error.

[0078] Specifically, in the embodiments of the present application, the speech recognition result detection apparatus judges whether the first evaluation result satisfies a preset condition, and obtains a first judgment result, including: comparing the first evaluation result with an evaluation threshold; in a case where the first evaluation result is less than the evaluation threshold, determining that the first judgment result is that the first evaluation result does not satisfy the preset condition; and in a case where the first evaluation result is greater than or equal to the evaluation threshold, determining that the first judgment result is that the first evaluation result does not satisfy the preset condition.

[0079] It can be understood that in the embodiments of the present application, the first evaluation result represents the overall score of the to-be-tested speech recognition result in terms of fluency, misspelling and the like, and therefore, in a case where the first evaluation result is less than the evaluation threshold, it indicates that the to-be-tested speech recognition result has an error, and correspondingly, in a case where the first evaluation result is greater than or equal to the evaluation threshold, it indicates that the to-be-tested speech recognition result has no error. The specific evaluation threshold can be set according to actual requirements, and the embodiments of the present application are not limited in this regard.

[0080] It should be noted that in the embodiments of the present application, the preset condition can be pre-set according to the actual error detection standard, and the specific preset condition is not limited in the embodiments of the present application. For example, the preset condition can also be a preset interval, that is, in the case that the first evaluation result exceeds the preset interval, it is determined that the error detection result of the to-be-tested speech recognition result is that there is an error, and in the case that the first evaluation result is within the preset interval, it is determined that the error detection result of the to-be-tested speech recognition result is that there is no error.

[0081] Figure 2 An exemplary error detection process of a speech recognition result is provided for the embodiments of the present application. As shown in Figure 2 , the speech recognition result detection device inputs the to-be-tested speech recognition result into the machine translation model, and the machine translation model translates the to-be-tested speech recognition result from a first language to a second language to output a to-be-tested translation result. Then, the translation score of the to-be-tested translation result, the perplexity score, and the language model score of the to-be-tested speech recognition result are obtained, so as to perform weighted summation by using the corresponding feature weights to obtain a first evaluation result. Finally, the first evaluation result is compared with an evaluation threshold to determine the error detection result of the to-be-tested speech recognition result.

[0082] It should be noted that in the embodiments of the present application, the speech recognition result detection device can actually determine the three feature weights in a certain manner before obtaining the three feature weights in the above step S103.

[0083] Figure 3 A flowchart for determining three feature weights is provided for the embodiments of the present application. As shown in Figure 3 , the following steps are mainly included:

[0084] S301, obtaining a sample speech recognition result and a preset detection result of the sample speech recognition result.

[0085] In the embodiments of the present application, the speech recognition result detection device can first obtain the sample speech recognition result and the preset detection result of the sample speech recognition result.

[0086] It should be noted that in the embodiments of the present application, the number of sample speech recognition results can be multiple. For example, several hundred thousand. The preset detection result of the sample speech recognition result is the judgment result of whether the sample speech recognition result has an error by a human being.

[0087] For example, in the embodiments of the present application, one sample speech recognition result is "First University 2019 New Student Opening", and its preset detection result is no error. In addition, another sample speech recognition result is "Second University Welcome 2019 New Student Opening", and its preset detection result is an error.

[0088] It should be noted that in the embodiments of the present application, a large number of sample speech recognition results can be obtained according to a certain proportion of preset detection results without errors and with errors, and the specific proportion is not limited in the embodiments of the present application.

[0089] It should be noted that in the embodiments of the present application, the sample speech recognition result and the to-be-detected speech recognition result are of the same language, i.e., the first language.

[0090] S302, the sample speech recognition result is translated from the first language to the second language by using the machine translation model to obtain a sample translation result.

[0091] In the embodiments of the present application, after obtaining the sample speech recognition result, the speech recognition result detection device can translate the sample speech recognition result from the first language to the second language by using the machine translation model to obtain a sample translation result.

[0092] It should be noted that in the embodiments of the present application, the process of translating the sample speech recognition result by the speech recognition result detection device using the machine translation model is the same as the process of translating the to-be-detected speech recognition result in the above step S101, and will not be repeated here.

[0093] S303, based on the sample translation result and the sample speech recognition result, a second translation score, a second perplexity score and a second language model score are determined.

[0094] In the embodiments of the present application, after obtaining the sample translation result and the sample speech recognition result, the speech recognition result detection device further determines a second translation score, a second perplexity score and a second language model score based on the sample translation result and the sample speech recognition result.

[0095] It should be noted that in the embodiments of the present application, the speech recognition result detection device determines a second translation score, a second perplexity score and a second language model score based on the sample translation result and the sample speech recognition result, which is similar to the process of determining a first translation score, a first perplexity score and a first language model score based on the to-be-detected translation result and the to-be-detected speech recognition result in the above step S102, and will not be repeated here.

[0096] S304, a preset weight corresponding to each of the second translation score, the second perplexity score and the second language model score is obtained to obtain three preset weights.

[0097] In the embodiments of the present application, the speech recognition result detection device can also obtain a preset weight corresponding to each of the second translation score, the second perplexity score and the second language model score to obtain three preset weights.

[0098] It should be noted that in the embodiments of the present application, because the importance of different types of features in the error detection process cannot be judged in the initial case, three preset weights can be set in advance, for example, each of the three preset weights can be set to 1. The three preset weights are not limited in the embodiments of the present application.

[0099] S305, determining three feature weights by using the second translation score, the second perplexity score, the second language model score, the preset detection result and the three preset weights.

[0100] In the embodiments of the present application, after the speech recognition result detection apparatus obtains the second translation score, the second perplexity score, the second language model score, the preset detection result and the three preset weights, the three feature weights can be determined by using the second translation score, the second perplexity score, the second language model score, the preset detection result and the three preset weights.

[0101] Specifically, in the embodiments of the present application, the speech recognition result detection apparatus determines the three feature weights by using the second translation score, the second perplexity score, the second language model score, the preset detection result and the three preset weights, including: performing weighted processing on the second translation score, the second perplexity score and the second language model score by using the three preset weights to obtain a second evaluation result; determining a second judgment result by judging whether the second evaluation result meets a preset condition, and determining an error detection result of the sample speech recognition result according to the second judgment result; adjusting the three preset weights based on the error detection result of the sample speech recognition result and the preset detection result to obtain the three feature weights.

[0102] It should be noted that in the embodiments of the present application, the speech recognition result detection apparatus performs weighted processing on the second evaluation feature by using the three preset weights, which is similar to the process of performing weighted processing on the first evaluation feature by using the three feature weights in the above step S103, and the difference is only the specific values of the features and weights, which will not be described here.

[0103] It should be noted that in the embodiments of the present application, the speech recognition result detection apparatus judges whether the second evaluation result meets the preset condition, which is similar to the process of judging whether the first evaluation result meets the preset condition in the above step S104, and the difference is only the object of judgment, which will not be described here.

[0104] Specifically, in the embodiments of the present application, the speech recognition result detection apparatus adjusts the three preset weights based on the error detection result of the sample speech recognition result and the preset detection result to obtain the three feature weights, including: adjusting the three preset weights according to a weight adjustment algorithm until the error detection result of the sample speech recognition result is the same as the preset detection result, to obtain the three feature weights.

[0105] It can be understood that, in the embodiment of the present application, the error detection result of the sample speech recognition result is the same as the preset detection result, that is, the weight setting is more appropriate, and therefore the three adjusted preset weights can be determined as the three feature weights.

[0106] It should be noted that, in the embodiment of the present application, the weight adjustment algorithm can be preset according to actual needs, such as a minimum error rate training (MERT) algorithm. The specific weight adjustment algorithm is not limited in the embodiment of the present application.

[0107] It can be understood that, in the embodiment of the present application, the speech recognition result detection device can determine the three feature weights by using a large number of sample speech recognition results, and therefore, in the weight adjustment process, when a high proportion of sample speech recognition results in the large number of sample speech recognition results have the same error detection result and preset detection result, the obtained weight can be determined as the feature weight.

[0108] It should be noted that, in the embodiment of the present application, the speech recognition result detection method described above can be applied to various application scenarios requiring speech recognition to achieve error detection of speech recognition results.

[0109] Figure 4 An exemplary system architecture diagram to which a speech recognition result detection method provided in an embodiment of the present application is applied is shown in FIG. 1. Figure 4 As shown in FIG. 1, the system can include a client, a cloud, a speech processing server, and a display screen, wherein the cloud is integrated with the speech recognition result detection method provided in the present application.

[0110] In actual application, in the process of a conference speech, the client collects speech data of a speaker, sends the collected speech data to the speech processing server, the speech processing server identifies the speech data to obtain a to-be-tested speech recognition result, then the speech processing server can send the to-be-tested speech recognition result to the cloud, the cloud performs error detection on the to-be-tested speech recognition result according to the speech recognition result detection method, and returns the error detection result to the speech processing server, if the error detection result is that there is an error, the speech processing server can correct the to-be-tested speech recognition result according to the error detection result in a certain way to obtain a correct speech recognition result, and finally the correct speech recognition result is projected onto the display screen for display.

[0111] It should be noted that, in the embodiment of the present application, the speech recognition result detection method described above is implemented in the form of software, which can be integrated not only in the cloud described above but also in a mobile terminal, and the embodiment of the present application is not limited in this regard.

[0112] The embodiment of the present application provides a speech recognition result detection method, comprising: obtaining a to-be-tested speech recognition result, and translating the to-be-tested speech recognition result from a first language to a second language by using a machine translation model to obtain a to-be-tested translation result; determining a first evaluation feature based on the to-be-tested translation result and the to-be-tested speech recognition result; the first evaluation feature is used for representing recognition and translation effects of the to-be-tested speech recognition result; evaluating the to-be-tested speech recognition result based on the first evaluation feature to obtain a first evaluation result; determining whether the first evaluation result meets a preset condition to obtain a first judgment result, and determining an error detection result of the to-be-tested speech recognition result according to the first judgment result. The technical scheme provided by the embodiment of the present application translates the to-be-tested speech recognition result into another language to combine the translation result to detect errors of the to-be-tested speech recognition, compared with detecting errors based on only a single language related feature, more features can be obtained to realize error detection, and the accuracy of error detection is improved.

[0113] The embodiment of the present application provides a speech recognition result detection device. Figure 5 The structure of the speech recognition result detection device provided by the embodiment of the present application is shown in Figure 1 . As shown in Figure 5 , the speech recognition result detection device comprises:

[0114] The translation module 501 is configured to obtain a to-be-tested speech recognition result, and translate the to-be-tested speech recognition result from a first language to a second language by using a machine translation model to obtain a to-be-tested translation result;

[0115] The determination module 502 is configured to determine a first evaluation feature based on the to-be-tested translation result and the to-be-tested speech recognition result; the first evaluation feature is used for representing recognition and translation effects of the to-be-tested speech recognition result;

[0116] The evaluation module 503 is configured to evaluate the to-be-tested speech recognition result based on the first evaluation feature to obtain a first evaluation result;

[0117] The judgment module 504 is configured to determine whether the first evaluation result meets a preset condition to obtain a first judgment result, and determine an error detection result of the to-be-tested speech recognition result according to the first judgment result.

[0118] In an embodiment, the determination module 502 is configured to obtain a translation score and a perplexity score of the machine translation model on the to-be-tested translation result to obtain a first translation score and a first perplexity score; input the to-be-tested speech recognition result into a language model corresponding to the first language to obtain a first language model score; and determine the first translation score, the first perplexity score and the first language model score as the first evaluation feature.

[0119] In an embodiment, the evaluation module 503 is configured to obtain a feature weight corresponding to each of the first translation score, the first perplexity score and the first language model score, to obtain three feature weights; and perform weighting processing on the first translation score, the first perplexity score and the first language model score by using the three feature weights, to obtain the first evaluation result.

[0120] In an embodiment, the determination module 502 is configured to obtain a sample speech recognition result and a preset detection result of the sample speech recognition result; translate the sample speech recognition result from the first language to the second language by using the machine translation model, to obtain a sample translation result; determine a second translation score, a second perplexity score and a second language model score based on the sample translation result and the sample speech recognition result; obtain a preset weight corresponding to each of the second translation score, the second perplexity score and the second language model score, to obtain three preset weights; and determine the three feature weights by using the second translation score, the second perplexity score, the second language model score, the preset detection result and the three preset weights.

[0121] In an embodiment, the determination module 502 is configured to perform weighting processing on the second translation score, the second perplexity score and the second language model score by using the three preset weights, to obtain a second evaluation result; determine whether the second evaluation result satisfies the preset condition, to obtain a second determination result, and determine an error detection result of the sample speech recognition result according to the second determination result; and adjust the three preset weights based on the error detection result of the sample speech recognition result and the preset detection result, to obtain the three feature weights.

[0122] In an embodiment, the determination module 502 is configured to adjust the three preset weights according to a weight adjustment algorithm until the error detection result of the sample speech recognition result is the same as the preset detection result, to obtain the three feature weights.

[0123] In an embodiment, the determination module 502 is configured to determine that the error detection result of the to-be-tested speech recognition result is erroneous in a case where the first determination result is that the first evaluation result does not satisfy the preset condition; and determine that the error detection result of the to-be-tested speech recognition result is error-free in a case where the first determination result is that the first evaluation result satisfies the preset condition.

[0124] In an embodiment, the determining module 504 is configured to compare the first evaluation result with an evaluation threshold value; in a case that the first evaluation result is less than the evaluation threshold value, determine that the first determination result is that the first evaluation result does not satisfy the preset condition; in a case that the first evaluation result is greater than or equal to the evaluation threshold value, determine that the first determination result is that the first evaluation result satisfies the preset condition.

[0125] It should be noted that, in actual application, the steps performed by the translation module 501, the determining module 502, the evaluation module 503 and the determining module 504 can be realized by a processor.

[0126] It should be noted that, in actual application, the steps performed by the translation module 501, the determining module 502, the evaluation module 503 and the determining module 504 can be realized by a processor.

[0127] Based on the hardware implementation of the above program modules, and in order to realize the method of the application embodiment, the application embodiment further provides a voice recognition result detection device. Figure 6 The structure of a voice recognition result detection device provided by the application embodiment Figure 2 As shown in FIG. 6, the voice recognition result detection device includes a processor 601, a memory 602 and a communication bus 603. Figure 6

[0128] The communication bus 603 is configured to realize the communication connection between the processor 601 and the memory 602.

[0129] The processor 601 is configured to execute the voice recognition result detection program stored in the memory 602, so as to realize the voice recognition result detection method described above.

[0130] ​The embodiment of the present application provides a speech recognition result detection device, obtains a to-be-detected speech recognition result, and translates the to-be-detected speech recognition result from a first language to a second language by using a machine translation model to obtain a to-be-detected translation result; determines a first evaluation feature based on the to-be-detected translation result and the to-be-detected speech recognition result; the first evaluation feature is used to represent the recognition and translation effect of the to-be-detected speech recognition result; the to-be-detected speech recognition result is evaluated based on the first evaluation feature to obtain a first evaluation result; it is judged whether the first evaluation result meets a preset condition to obtain a first judgment result, and the error detection result of the to-be-detected speech recognition result is determined according to the first judgment result. The speech recognition result detection device provided by the embodiment of the present application translates the to-be-detected speech recognition result into another language, so as to detect errors of the to-be-detected speech recognition by combining the translation result. Compared with the error detection based on only a single language related feature, more features can be obtained to realize error detection, and the accuracy of error detection is improved.

[0131] The embodiment of the present application also provides a computer readable storage medium, which stores a computer program. The computer program is executed by one or more processors to implement the above-mentioned simultaneous interpretation method. The computer readable storage medium can be a volatile memory (volatile memory), such as random access memory (Random-Access Memory, RAM); or a non-volatile memory (non-volatile memory), such as read-only memory (Read-Only Memory, ROM), flash memory (flash memory), hard disk (Hard Disk Drive, HDD) or solid state disk (Solid-State Drive, SSD); or a device including one or any combination of the above memories, such as a mobile phone, a computer, a tablet device, a personal digital assistant, etc.

[0132] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can be in the form of a hardware embodiment, a software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can be in the form of a computer program product implemented on one or more computer usable storage media (including but not limited to magnetic disk storage and optical storage, etc.) containing computer usable program code.

[0133] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart or flowsheet block or blocks. Figure 1 one or more functions specified in the flowchart or flowsheet block or blocks. Figure 1

[0134] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart or flowsheet block or blocks. Figure 1 one or more functions specified in the flowchart or flowsheet block or blocks. Figure 1

[0135] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart or flowsheet block or blocks. Figure 1 one or more functions specified in the flowchart or flowsheet block or blocks. Figure 1

[0136] The above generally describes only some embodiments of the present application, and the protection scope of the present application is not limited thereto. Any changes or replacements easily conceived by those skilled in the art within the technical scope disclosed by the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.​​​

Claims

1. A method for detecting speech recognition results, comprising: Obtaining a speech recognition result to be tested, and translating the speech recognition result to be tested from a first language to a second language using a machine translation model to obtain a translation result to be tested; Determining a first evaluation feature based on the translation result to be tested and the speech recognition result to be tested; The first evaluation feature is used to characterize the recognition and translation effects of the speech recognition result to be tested; Evaluate the speech recognition result to be tested based on the first evaluation feature to obtain a first evaluation result; Determining whether the first evaluation result meets a preset condition to obtain a first determination result, and determining an error detection result of the speech recognition result to be tested based on the first determination result; Wherein, the first evaluation feature includes a first translation score and a first perplexity score of the translation result to be tested; The first translation score and the first perplexity score are results of considering the recognition effect of the speech recognition result to be tested based on the language features of the second language; The first translation score represents the overall translation effect of the translation result to be tested, and the first perplexity score represents the degree of confusion encountered by the translation result to be tested during the translation process.

2. The method according to claim 1, wherein The determining of a first evaluation feature based on the translation result to be tested and the speech recognition result to be tested includes: Obtaining a translation score and a perplexity score of the machine translation model for the translation result to be tested, to obtain a first translation score and a first perplexity score; Inputting the speech recognition result to be tested into a language model corresponding to the first language to obtain a first language model score; The first translation score, the first perplexity score, and the first language model score are determined as the first evaluation features.

3. The method according to claim 2, wherein: The step of evaluating the speech recognition result to be tested based on the first evaluation feature to obtain a first evaluation result includes: Obtaining a feature weight corresponding to each of the first translation score, the first perplexity score, and the first language model score to obtain three feature weights; The first translation score, the first perplexity score, and the first language model score are weighted using the three feature weights to obtain the first evaluation result.

4. The method according to claim 3, wherein: Before obtaining the feature weight corresponding to each of the first translation score, the first perplexity score, and the first language model score to obtain the three feature weights, the method further includes: Obtaining a sample speech recognition result and a preset detection result of the sample speech recognition result; translating the sample speech recognition result from the first language into the second language using the machine translation model to obtain a sample translation result; Determining a second translation score, a second perplexity score, and a second language model score based on the sample translation result and the sample speech recognition result; Obtaining a preset weight corresponding to each of the second translation score, the second perplexity score, and the second language model score to obtain three preset weights; The three feature weights are determined using the second translation score, the second perplexity score, the second language model score, the preset detection result, and the three preset weights.

5. The method according to claim 4, wherein The determining the three feature weights by using the second translation score, the second perplexity score, the second language model score, the preset detection result, and the three preset weights includes: The second translation score, the second perplexity score, and the second language model score are weighted using the three preset weights to obtain a second evaluation result; the second translation score, the second perplexity score, and the second language model score are used to represent the recognition and translation effect of the sample speech recognition result; Determining whether the second evaluation result meets the preset condition to obtain a second determination result, and determining an error detection result of the sample speech recognition result according to the second determination result; Based on the error detection result of the sample speech recognition result and the preset detection result, the three preset weights are adjusted to obtain the three feature weights.

6. The method according to claim 5, wherein: The adjusting of the three preset weights based on the error detection result of the sample speech recognition result and the preset detection result to obtain the three feature weights includes: The three preset weights are adjusted according to a weight adjustment algorithm until the error detection result of the sample speech recognition result is the same as the preset detection result, thereby obtaining the three feature weights.

7. The method according to any one of claims 1 to 6, wherein: The step of determining the error detection result of the speech recognition result to be tested according to the first judgment result includes: If the first judgment result is that the first evaluation result does not meet the preset condition, determining that the error detection result of the speech recognition result to be tested is an error; In a case where the first judgment result is that the first evaluation result satisfies the preset condition, it is determined that the error detection result of the speech recognition result to be tested is error-free.

8. The method according to any one of claims 1 to 6, wherein: The determining whether the first evaluation result satisfies a preset condition to obtain a first determination result includes: comparing the first evaluation result with an evaluation threshold; If the first evaluation result is less than the evaluation threshold, determining the first judgment result is that the first evaluation result does not meet the preset condition; In a case where the first evaluation result is greater than or equal to the evaluation threshold, a first judgment result is determined as that the first evaluation result meets the preset condition.

9. A speech recognition result detection device, comprising: A translation module is configured to obtain a speech recognition result to be tested, and translate the speech recognition result to be tested from a first language to a second language using a machine translation model to obtain a translation result to be tested; a determination module configured to determine a first evaluation feature based on the translation result to be tested and the speech recognition result to be tested; The first evaluation feature is used to characterize the recognition and translation effects of the speech recognition result to be tested; an evaluation module configured to evaluate the speech recognition result to be tested based on the first evaluation feature to obtain a first evaluation result; a judgment module configured to judge whether the first evaluation result meets a preset condition, obtain a first judgment result, and determine an error detection result of the speech recognition result to be tested according to the first judgment result; Wherein, the first evaluation feature includes a first translation score and a first perplexity score of the translation result to be tested; The first translation score and the first perplexity score are results of considering the recognition effect of the speech recognition result to be tested based on the language features of the second language; The first translation score represents the overall translation effect of the translation result to be tested, and the first perplexity score represents the degree of confusion encountered by the translation result to be tested during the translation process.

10. A speech recognition result detection device, the device comprising a processor and a memory; The processor is configured to execute the speech recognition result detection program stored in the memory to implement the speech recognition result detection method according to any one of claims 1 to 8.

11. A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the method for detecting speech recognition results according to any one of claims 1 to 8 is implemented.

Images