A speech evaluation method, device, electronic device and storage medium

Abstract

This application provides a speech evaluation method, apparatus, electronic device, and storage medium. The speech evaluation method comprises: obtaining a reading speech generated by a subject reading a standard text and determining the effective speech duration of the reading speech; obtaining the reading accuracy of the reading speech based on the reading speech and the standard text; and generating a fluency evaluation result for the reading speech based on the effective speech duration and the reading accuracy of the reading speech. This method evaluates the reading speech by combining two different dimensions of speech data: reading accuracy and effective speech duration. This improves both the efficiency of speech evaluation and the accuracy of the speech evaluation results.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and specifically to a speech evaluation method, device, electronic device, and storage medium. Background Art

[0002] Reading ability is an important factor in measuring learning ability. In the prior art, the rapid expression ability of the subject is usually examined based on the rapid naming task of text content such as numbers, letters and words, and the fluency of the subject's speech when completing the rapid naming task is evaluated. The more fluent the subject's reading speech is, the stronger the subject's reading ability is, and vice versa. However, in the evaluation stage, the determination of speech fluency mostly relies on manual recording. This method not only consumes a lot of human resources, but also has low efficiency in speech evaluation, and the accuracy of speech evaluation cannot be guaranteed.

[0003] Therefore, how to evaluate speech to improve speech evaluation efficiency and speech evaluation accuracy has become a technical problem that those skilled in the art urgently need to solve. Summary of the Invention

[0004] The present application provides a speech evaluation method, apparatus, electronic device, and storage medium to improve speech evaluation efficiency and accuracy.

[0005] According to a first aspect of an embodiment of the present application, a speech evaluation method is provided, comprising:

[0006] Obtaining a reading speech generated by the subject reading a standard text, and determining an effective speech duration of the reading speech;

[0007] Obtaining a reading accuracy of the reading speech according to the reading speech and the standard text;

[0008] A fluency evaluation result for the reading speech is generated based on the effective speech duration of the reading speech and the reading accuracy of the reading speech.

[0009] In an optional implementation manner of the present application, determining the effective duration of the reading speech includes:

[0010] The standard text and the reading voice are aligned to determine the effective duration of the reading voice.

[0011] In an optional implementation manner of the present application, obtaining the reading accuracy of the reading speech according to the reading speech and the standard text includes:

[0012] Obtaining reading error information in the reading speech according to the reading speech and the standard text; the reading error information includes at least one of phoneme reading error information and text reading error information;

[0013] Obtaining the reading accuracy of the reading speech according to the reading error information; or obtaining the reading accuracy of the reading speech according to the reading error information and feature information of the reading speech;

[0014] The characteristic information of the reading voice includes at least one of the recognized text of the reading voice and the effective voice duration of the reading voice.

[0015] In an optional implementation manner of the present application, obtaining reading error information in the reading speech according to the reading speech and the standard text includes:

[0016] The reading speech and the standard text are input into a pre-trained error detection model, so that the error detection model performs error detection processing on the reading speech based on the standard text to obtain reading error information in the reading speech.

[0017] In an optional implementation manner of the present application, obtaining the reading accuracy of the reading speech according to the reading speech and the standard text includes:

[0018] Performing speech recognition processing on the reading speech to obtain a recognition text corresponding to the reading speech;

[0019] Determining a common text sequence between the recognition text and the standard text;

[0020] The reading accuracy of the reading speech is obtained according to the common text sequence and the standard text.

[0021] In an optional implementation manner of the present application, obtaining the reading accuracy of the reading speech according to the common text sequence and the standard text includes:

[0022] Obtaining a first score for the reading speech according to a sequence length of the common text sequence;

[0023] Inputting the reading speech and the standard text into a pre-trained error detection model, so that the error detection model performs error detection processing on the reading speech based on the standard text, and obtains text reading error information in the reading speech;

[0024] Obtaining a second score for the reading speech according to the text reading error information;

[0025] Performing linear regression on the first score and the second score to obtain the reading accuracy of the reading speech.

[0026] In an optional embodiment of the present application, after performing speech recognition processing on the reading speech to obtain a recognized text corresponding to the reading speech, the method further includes:

[0027] Inputting the reading speech and the recognized text into a pre-trained error detection model, so that the error detection model performs error detection processing on the reading speech based on the recognized text, and obtains text reading error information in the reading speech;

[0028] Based on the text reading error information, the recognized text is subjected to denoising processing to obtain a recognized text after denoising processing, wherein the recognized text after denoising processing includes a correctly read text sequence;

[0029] The determining of a common text sequence between the recognition text and the standard text includes:

[0030] Determine a common text sequence between the recognized text after denoising and the standard text.

[0031] In an optional embodiment of the present application, the error detection model is trained in the following manner:

[0032] Obtaining the first sample speech with standard reading error information;

[0033] The first sample speech is used as a training sample to perform error detection training on a pre-built error detection model.

[0034] In an optional embodiment of the present application, generating a fluency evaluation result for the reading speech based on the effective speech duration of the reading speech and the reading accuracy of the reading speech includes:

[0035] Generating a penalty duration in a time dimension according to the reading accuracy;

[0036] The penalty duration is added to the effective duration of the reading speech to generate a fluency evaluation result for the reading speech.

[0037] In an optional implementation manner of the present application, before obtaining the reading accuracy of the reading speech based on the reading speech and the standard text, the method further includes:

[0038] Inputting the reading speech into a pre-trained binary classification model so that the binary classification model scores the reading speech and determines whether the score of the reading speech reaches a preset scoring threshold;

[0039] If the score of the reading speech does not reach the preset score threshold, performing the step of obtaining the reading accuracy of the reading speech based on the reading speech and the standard text;

[0040] If the score of the reading speech reaches a preset score threshold, a fluency evaluation result for the reading speech is generated based on the effective speech duration of the reading speech.

[0041] In an optional embodiment of the present application, the binary classification model is obtained by:

[0042] Obtaining a second sample speech labeled with a score;

[0043] The second sample speech is used as a training sample to perform scoring training on the pre-built binary classification model.

[0044] According to a second aspect of an embodiment of the present application, a speech evaluation device is provided, comprising:

[0045] The first unit is used to obtain the reading speech generated by the subject reading the standard text and determine the effective speech duration of the reading speech;

[0046] The second unit is configured to obtain a reading accuracy of the reading speech according to the reading speech and the standard text;

[0047] The third unit is configured to generate a fluency evaluation result for the reading speech based on the effective speech duration of the reading speech and the reading accuracy of the reading speech.

[0048] According to a third aspect of the embodiments of the present application, there is provided an electronic device, including:

[0049] processor;

[0050] a memory for storing instructions executable by the processor;

[0051] The processor is configured to execute the above-mentioned speech evaluation method by running the instructions in the memory.

[0052] According to a fourth aspect of the embodiments of the present application, a computer-readable storage medium is provided, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned speech evaluation method is executed.

[0053] Compared with the prior art, this application has the following advantages:

[0054] The speech evaluation method, apparatus, electronic device, and storage medium provided in this application, after obtaining a reading speech, determines the reading accuracy of the reading speech based on the reading speech and a standard text, and then evaluates the reading speech based on the reading accuracy and the effective speech duration of the reading speech. This method abandons the traditional method of manual recording and evaluates the reading speech by combining two different dimensions of speech data: reading accuracy and effective speech duration. This method not only improves the efficiency of speech evaluation, but also improves the accuracy of the speech evaluation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without any creative work.

[0056] Figure 1 A schematic diagram of a question book for a rapid number naming task provided in one embodiment of the present application;

[0057] Figure 2 A schematic diagram of an application scenario of a speech evaluation method provided in another embodiment of the present application;

[0058] Figure 3 A flow chart of a speech evaluation method provided in another embodiment of the present application;

[0059] Figure 4 A schematic diagram of the structure of a speech evaluation device provided in another embodiment of the present application;

[0060] Figure 5 A schematic diagram of the structure of an electronic device provided in another embodiment of the present application. DETAILED DESCRIPTION

[0061] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0062] Reading ability is an important factor in measuring learning ability. In the prior art, the rapid expression ability of the subject is usually examined based on the rapid naming task of text content such as numbers, letters and words, and the fluency of the subject's speech when completing the rapid naming task is evaluated. The more fluent the subject's reading speech is, the stronger the subject's reading ability is, and vice versa. However, in the evaluation stage, the determination of speech fluency mostly relies on manual recording. This method not only consumes a lot of human resources, but also has low efficiency in speech evaluation, and the accuracy of speech evaluation cannot be guaranteed.

[0063] Therefore, how to evaluate speech to improve speech evaluation efficiency and speech evaluation accuracy has become a technical problem that those skilled in the art urgently need to solve.

[0064] In order to solve the above technical problems, the present application provides a speech evaluation method, device, electronic device and storage medium, which will be described in detail one by one in the following embodiments.

[0065] Exemplary Implementation Environment

[0066] First, in order to facilitate understanding of the specific application scenario of the speech evaluation method provided in this application, the speech evaluation method is introduced below in conjunction with an application scenario of the speech evaluation method.

[0067] The embodiment of the present application scenario specifically describes a process of evaluating the reading voice of a subject when reading a quick naming task question book after the subject completes the quick naming task.

[0068] Furthermore, in order to facilitate understanding of the rapid naming task, the task is first introduced in conjunction with a rapid naming task question book.

[0069] Please refer to Figure 1 , Figure 1 A schematic diagram of a question book for a rapid number naming task provided in one embodiment of the present application.

[0070] like Figure 1 As shown, Figure 1 The question book shown includes fifty numbers. In the rapid naming task, the subject is required to read these numbers accurately from left to right at the fastest speed based on the visual information of the question book, so as to obtain the reading voice of the subject when reading the question book and evaluate the reading voice.

[0071] Understandably, Figure 1The question book shown is only an optional question book for the quick naming task. In actual applications, the question book for the quick naming task can also have other forms. For example, the question book for the quick naming task can also record letters, words, or mathematical symbols, etc. This application does not impose any restrictions on this.

[0072] For further information, please refer to Figure 2 , Figure 2 A schematic diagram of an application scenario of a speech evaluation method provided in another embodiment of the present application.

[0073] Figure 2 The system includes: a measured object 201, a voice collection device 202, and a scoring device 203.

[0074] The object under test 201 can be understood as Figure 1 The question book shown is for a person performing a number naming task. In some other optional embodiments of the present application, the test object 201 may also be a terminal device such as an artificial intelligence device, a speech synthesis device, or an intelligent robot. For example, the intelligent robot may be a chat robot capable of emitting sound information, a navigation robot capable of giving directions by voice, and the like. When the test object 201 is an intelligent robot, the reading voice of the intelligent robot may be obtained when the intelligent robot recognizes and reads the question book, and the fluency of the reading voice may be evaluated. In an optional embodiment of the present application, the text recognition ability of the intelligent robot may also be evaluated in combination with the fluency evaluation result of the reading voice.

[0075] The voice collection device 202 can be understood as a device for collecting the reading voice of the subject 201 when reading the question book. The voice collection device 202 can be a terminal device such as a voice sensor, a recorder, or a computer.

[0076] Scoring device 203 can be understood as a scoring device based on the reading voice and Figure 1 The question book shown is a tool for evaluating the fluency of the reading speech, wherein the scoring device 203 performs the following steps S21 to S29 during the speech evaluation process:

[0077] Step S21, start.

[0078] Step S22: obtaining reading speech.

[0079] This step can be understood as obtaining the voice information of the subject 201 reading the question book collected by the voice collection device 202. After obtaining the reading voice, the following step S23 is further performed.

[0080] Step S23: input the reading speech into a binary classification model to score the reading speech and determine whether the reading speech is a full-score speech.

[0081] If the reading voice is a full-score voice, step S24 is executed;

[0082] Step S24: input the reading speech into an error detection model, obtain the effective speech duration of the reading speech output by the error detection model, and use the effective speech duration as a fluency evaluation result of the reading speech.

[0083] The error detection model includes an alignment module for aligning the reading speech with the question book to obtain the effective speech duration of the reading speech. After the reading speech is scored by the binary classification model, if the reading speech receives a full score, the effective speech duration of the reading speech is used as the evaluation result of the reading speech fluency, and step S29 is executed.

[0084] That is, if the reading voice is a full-score voice, it can be considered that the reading accuracy of the reading voice is 100%. At this time, the effective voice duration of the reading voice can be used as the evaluation result of the reading voice fluency.

[0085] If the reading voice is a non-full score video, then execute steps S25 and S26;

[0086] Step S25 , inputting the reading speech into an error detection model, and obtaining the effective speech duration of the reading speech and reading error information of the reading speech output by the error detection model.

[0087] Step S26: performing speech recognition on the reading speech to obtain a recognized text of the reading speech.

[0088] Among them, on the one hand, the error detection model is used to perform error detection processing on the reading speech to obtain reading error information in the reading speech, and the reading error information includes: phoneme reading error information and text reading error information; on the other hand, the error detection model contains an alignment module for aligning the reading speech and the question book to obtain the effective speech duration of the reading speech.

[0089] After executing the above steps S25 and S26 to obtain the effective voice duration, the reading error information of the reading voice, and the recognized text, step S27 is further executed.

[0090] Step S27 , obtaining the reading accuracy of the reading speech according to the effective speech duration, the reading error information, and the recognized text.

[0091] The above step S27 can be implemented by linear regression. After obtaining the reading accuracy, step S28 is executed.

[0092] Step S28: generating a fluency evaluation result of the reading speech according to the reading accuracy and the effective speech duration.

[0093] Specifically, the above step S28 can be understood as generating a penalty duration in the time dimension according to the reading accuracy; and adding the penalty duration on the basis of the effective voice duration of the reading voice to generate a fluency evaluation result for the reading voice.

[0094] Step S29, end.

[0095] It can be understood that the above introduction to the embodiment of the scenario of this application is only to facilitate a better understanding of the speech evaluation method provided by this application, and is not used to limit the application scenario of the speech evaluation method. The speech recognition method can also be applied to other scenarios, for example, for speech evaluation of artificial intelligence robots or speech evaluation of character dubbing, etc.

[0096] Exemplary Methods

[0097] In an exemplary embodiment of the present application, a speech evaluation method is provided, the core of which is to evaluate the reading speech by combining two different dimensions of speech data, namely, the reading accuracy and the effective speech duration of the reading speech, so as to improve the speech evaluation efficiency and the accuracy of the speech evaluation results.

[0098] In an optional embodiment of the present application, the implementation subject of the voice evaluation method can be various types of user terminals such as laptops, tablet computers, desktop computers, set-top boxes, mobile devices (for example: mobile phones, portable music players, personal digital assistants, dedicated messaging devices, recorders, game consoles), or a combination of any two or more of these data processing devices, or a server.

[0099] Please refer to Figure 3 , Figure 3 This is a flow chart of a speech evaluation method provided in another embodiment of the present application.

[0100] like Figure 3 As shown, the voice evaluation method includes the following steps S301 to S303:

[0101] Step S301: obtaining a reading speech generated by a test subject reading a standard text, and determining an effective speech duration of the reading speech.

[0102] In an optional embodiment of the present application, the standard text read by the subject can be as follows: Figure 1 The question book shown for the quick naming task can also be a text file in other forms, such as: an article, a string of English words, etc., and this application does not impose any limitation on this.

[0103] The test object can be understood as a person who reads the standard text, or a terminal device such as an artificial intelligence device, a speech synthesis device, or an intelligent robot. For example, the intelligent robot can be a chat robot that can emit voice information, a navigation robot that can provide voice directions, etc. When the test object 201 is an intelligent robot, the reading speech of the intelligent robot obtained when the intelligent robot recognizes and reads the question book can be used to evaluate the fluency of the reading speech. In an optional embodiment of the present application, the text recognition ability of the intelligent robot can also be evaluated in combination with the fluency evaluation results of the reading speech.

[0104] Furthermore, the reading voice can be acquired through a pre-prepared sound sensor. For example, the sound sensor can be a mobile phone, computer, voice recorder, etc. of the subject or a staff member serving the current scene, or a terminal device such as a recorder or microphone set up in the scene where the subject is located and dedicated for voice acquisition.

[0105] In the embodiment of the present application, the effective speech duration of the reading speech can be understood as the total time spent from the first symbol to the last symbol of the standard text when the subject performs the reading task. For example, assuming that the standard text is Figure 1 For the question book of the rapid number naming task shown in the figure, the effective speech duration of the reading speech refers to the total time spent by the subject reading the first number 5 in the first line of the question book to the last number 4 in the last line.

[0106] In the process of actual application, in the process of collecting reading voice, some other content that is not related to the standard text may be collected. For example, in the process of collecting reading voice for the rapid naming task, the subject may determine the starting time point of the rapid naming task, and then before the subject starts reading the standard text, he may make a question-like voice such as "Can we start?" For another example, assuming that there is a host in the environment where the subject is located, the host's voice may also appear in the reading voice before or after the subject reads the standard text. For another example, in the process of collecting reading voice, because the standard text is not ready, there may be no sound in the reading voice.

[0107] In an optional implementation manner of the present application, in order to eliminate the influence of the above content on the speech evaluation result, determining the effective speech duration of the reading speech includes:

[0108] The standard text and the reading speech are aligned to determine the effective speech duration of the reading speech.

[0109] In an optional embodiment of the present application, the alignment process of the standard text and the reading speech can be understood as aligning each phoneme in the reading speech with each byte in the standard text, thereby obtaining the effective duration of the reading speech.

[0110] In another optional embodiment of the present application, in order to more accurately determine the effective voice duration of the reading voice, the determination of the effective voice duration of the reading voice includes: performing voice boundary detection on the reading voice to obtain a first audible segment of the reading voice; aligning the standard text with the reading voice to obtain a second audible segment of the reading voice; obtaining an intersection segment of the first audible segment and the second audible segment, and determining the duration of the intersection segment as the effective voice duration of the reading voice.

[0111] Furthermore, considering that in the process of collecting the reading voice, different voice collection devices collect different voice formats. Therefore, before executing the above step S301, it is also necessary to normalize the voice format of the reading voice, so that the reading voice can be smoothly evaluated and processed later. For example, in some scenarios, the reading voice may be recorded through the web page. At this time, the format of the reading voice is 16k*8bit, but in the subsequent processing of the reading voice, the format of the reading voice is required to be 16k*16bit. At this time, the 16k*8bit reading voice needs to be converted into a reading voice with a format of 16k*16bit.

[0112] Specifically, the above-mentioned process of normalizing the voice format of the reading voice can be implemented based on a preset voice format conversion software application.

[0113] Furthermore, during the recording of the reading speech, the reading speech may contain a significant amount of noise due to the influence of direct current (DC) in the recording device. Therefore, to further improve the quality of the reading speech and the accuracy of the reading speech evaluation, the method further includes: performing a denoising process on the reading speech by removing DC.

[0114] Step S302 : obtaining the reading accuracy of the reading speech according to the reading speech and the standard text.

[0115] The accuracy of the reading speech can be understood as the accuracy rate of the measured object when reading the standard text. In an optional embodiment of the present application, the above step S302 can be implemented by the following steps S1 and S2:

[0116] Step S1: obtaining reading error information in the reading speech according to the reading speech and the standard text.

[0117] The reading error information includes at least one of phoneme reading error information and text reading error information.

[0118] In an optional implementation manner of the present application, the above step S1 includes:

[0119] The reading speech and the standard text are input into a pre-trained error detection model, so that the error detection model performs error detection processing on the reading speech based on the standard text to obtain reading error information in the reading speech.

[0120] Among them, the error detection model can be understood as a convolutional neural network. In the process of specific application, the present application adopts machine learning (ML) to train and obtain the error detection model. Machine learning (a multi-disciplinary interdisciplinary subject involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines) is dedicated to studying the acquisition of new knowledge or skills through training samples, reorganizing existing knowledge structures and continuously improving its own performance. Machine learning usually includes artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning and other technologies, and is a branch of artificial intelligence (AI) technology. In an optional embodiment of the present application, the error detection model can be obtained by training sample voices marked with reading error information. Step S2, according to the reading error information, obtain the reading accuracy of the reading voice; or, according to the reading error information and the feature information of the reading voice, obtain the reading accuracy of the reading voice.

[0121] The characteristic information of the reading voice includes at least one of the recognized text of the reading voice and the effective voice duration of the reading voice.

[0122] Specifically, the above step S2 includes performing linear regression on at least one of the text feature of the recognized text and the duration feature of the effective voice duration, and the error feature of the reading error information to obtain the reading accuracy.

[0123] In an optional embodiment of the present application, the reading accuracy can be denoted as S, the text feature of the recognized text can be denoted as x, the duration feature of the effective speech duration can be denoted as y, and the error feature of the reading error can be denoted as z. Then, the regression equation for linear regression can be expressed by the following formula (1):

[0124] S=Ax+By+Cz+D (1)

[0125] Among them, A, B, C are regression coefficients, and D is a constant.

[0126] In another optional implementation of the present application, the above step S302 can also be implemented by the following steps S3 to S5:

[0127] Step S3: performing speech recognition processing on the reading speech to obtain a recognition text corresponding to the reading speech.

[0128] Specifically, the process of performing speech recognition processing on the reading speech can be implemented using a preset speech recognition model. Similar to the error detection model, the speech recognition model is also a convolutional neural network. Furthermore, the speech recognition model can be obtained by training a pre-built speech recognition model using sample speech annotated with text identifiers.

[0129] Step S4: determining a common text sequence between the recognition text and the standard text.

[0130] Specifically, the common text sequence of the recognition text and the standard text can be understood as an overlapping character string of the recognition text and the standard text.

[0131] For example, assuming that the standard text is "1, 2, 3, 4, 5, 6", and based on the speech recognition processing of the reading speech, the recognized text obtained is "1, 2, 3, 4, 7, 6", then the common text sequence is "1, 2, 3, 4, 6".

[0132] Step S5: obtaining the reading accuracy of the reading speech according to the common text sequence and the standard text.

[0133] In an optional implementation manner of the present application, the above step S5 includes the following sub-steps S5-1 to S5-4:

[0134] Step S5-1: Obtain a first score for the reading speech according to the sequence length of the common text sequence.

[0135] The first score of the reading speech can be understood as the score after removing the reading errors from the reading speech. For example, assuming that the full score of the reading accuracy is 50 points and there are 50 characters in the standard text, if the sequence length of the common text sequence is 49, the first score can be 49 points.

[0136] In an optional embodiment of the present application, the above step S5-1 includes: determining a score coefficient for each character of the standard text; and multiplying the sequence length of the common text sequence by the score coefficient to obtain a first score for the reading speech.

[0137] In another optional embodiment of the present application, the above-mentioned step S5-1 includes: aligning the standard text and the common text sequence to determine the length of the aligned text sequence; determining the score coefficient of each character in the text; and multiplying the length of the aligned text sequence by the score coefficient to obtain a first score for the reading speech.

[0138] Step S5-2: input the reading speech and the standard text into a pre-trained error detection model, so that the error detection model performs error detection processing on the reading speech based on the standard text, and obtains text reading error information in the reading speech.

[0139] Step S5-3: Obtain a second score for the reading speech according to the text reading error information.

[0140] In an embodiment of the present application, the second score of the reading speech can also be understood as the score after removing the reading error part from the reading speech. The difference from the first score is that the process of removing errors for the reading speech is implemented based on text reading error information, and the text reading error information is obtained based on the error detection model, that is, through the error detection model, the text reading error information in the reading speech is obtained, and the number of characters read incorrectly in the reading speech is determined. Afterwards, based on the total number of characters in the reading text, the number of characters read correctly in the reading speech is determined.

[0141] For example, assuming that the full score of the reading accuracy is 50 points, and there are 50 characters in the standard text, if the error detection model detects a total of two text reading error messages, the second score may be 48 points.

[0142] Step S5-4: performing linear regression processing on the first score and the second score to obtain the reading accuracy of the reading speech.

[0143] Furthermore, in an optional embodiment of the present application, the first score may be s1, and the second score may be s2. Then, the regression equation for linear regression may be expressed by the following formula (2):

[0144] S=As1+Bs2+C (2)

[0145] Among them, A and B are regression coefficients, and C is a constant.

[0146] In another optional implementation manner of the present application, the accuracy of the reading speech may also directly adopt the above-mentioned first score or the second score, and the present application does not impose any limitation on this.

[0147] In an optional embodiment of the present application, in order to determine the accuracy of the reading speech, after the above step S3, the method further includes the following steps S6 and S7:

[0148] Step S6: input the reading speech and the recognized text into a pre-trained error detection model, so that the error detection model performs error detection processing on the reading speech based on the recognized text, and obtains text reading error information in the reading speech.

[0149] Different from the application process of the error detection model in the above-mentioned steps S1 and S5-2, in step S6, the input data of the error detection model is the reading speech and the recognition text corresponding to the reading speech, and the output data is the text reading information after the reading speech is error-detected based on the recognition text of the reading speech.

[0150] Here, the embodiment of the present application uses the recognition text corresponding to the reading speech as the error detection standard for the reading speech because the speech recognition process is based on the similarity between the pronunciation characteristics of each phoneme in the speech and the pronunciation characteristics of the standard. Therefore, during the speech recognition process, it is possible that the recognized text corresponds to the standard text, but the actual pronunciation in the speech is inaccurate or erroneous.

[0151] For example, assuming that the recognized text obtained after recognizing the reading voice is "1, 2, 3, 4, 7, 6", after inputting the recognized text and the reading voice into the error detection model, if the 4 in the reading voice is erroneous, the error detection model outputs the text reading error information of 4 bytes.

[0152] Step S7: Based on the text reading error information, the recognized text is subjected to denoising processing to obtain a recognized text after denoising processing, wherein the recognized text after denoising processing includes a correctly read text sequence.

[0153] In the embodiment of the present application, the denoising process for the recognized text based on the text reading error information refers to removing bytes in the recognized text corresponding to the text error information based on the text reading error information.

[0154] For example, if the recognized text is "1, 2, 3, 4, 7, 6" and the output of the error detection model indicates that there is text reading error information in byte 4, then the recognized text after denoising is "1, 2, 3, -, 7, 6".

[0155] Furthermore, after denoising the recognized text to obtain the denoised recognized text, it is necessary to further determine the obvious errors in the recognized text. For example, assuming the standard text is "1, 2, 3, 4, 5, 6", and the denoised recognized text is "1, 2, 3, -, 7, 6", it can be determined that character 7 in the recognized text is incorrect.

[0156] In an optional implementation of the present application, after executing step S7, step S4 may be continued to be executed, wherein step S4 includes: determining a common text sequence between the recognized text after denoising processing and the standard text.

[0157] Furthermore, in an optional embodiment of the present application, in the process of determining the common text sequence between the recognized text after denoising and the standard text, error detection of the reading speech based on the error detection model and recognition text based on the standard text have been taken into consideration.

[0158] Therefore, after obtaining the common text sequence between the recognized text and the standard text, the reading accuracy of the reading speech can be determined based on the sequence length of the common text sequence and the score coefficient of each character; the common text sequence and the standard text can also be further aligned, and the reading accuracy of the reading speech can be determined based on the number of aligned characters and the score coefficient of each character. This application does not impose any restrictions on this.

[0159] Step S303 : generating a fluency evaluation result for the reading speech based on the effective speech duration of the reading speech and the reading accuracy of the reading speech.

[0160] The process evaluation result of the reading speech can be understood as an evaluation of the reading time and reading accuracy of the reading speech. In an optional embodiment of the present application, the above step S303 includes the following steps S8 to S9:

[0161] Step S8: Generate a penalty duration in the time dimension according to the reading accuracy.

[0162] Step S9 , adding the penalty duration to the effective duration of the reading speech to generate a fluency evaluation result for the reading speech.

[0163] In an optional implementation of the present application, the fluency evaluation result of the reading speech can be expressed by the following formula (3):

[0164] T=T0+|MS|*q (3)

[0165] Among them, T represents the fluency evaluation result of the reading speech, T0 represents the effective speech duration of the reading speech; M represents the reading accuracy score of 100%; S represents the reading accuracy score of the reading speech, q represents the penalty coefficient of the time dimension; |MS|*q represents the penalty duration of the time dimension.

[0166] In an optional implementation of the present application, considering the evaluation efficiency problem in large-scale speech evaluation, before executing the above step S302, the speech evaluation method further includes the following steps S304 to S306:

[0167] Step S304 : inputting the reading speech into a pre-trained binary classification model, so that the binary classification model scores the reading speech and determines whether the score of the reading speech reaches a preset scoring threshold.

[0168] Step S305 : If the score of the reading speech does not reach the preset score threshold, the step of obtaining the reading accuracy of the reading speech based on the reading speech and the standard text is performed.

[0169] Step S306: If the score of the reading speech reaches a preset score threshold, a fluency evaluation result for the reading speech is generated based on the effective speech duration of the reading speech.

[0170] In an optional implementation of the present application, the score of the reading speech may be a reading accuracy score of the reading speech, and accordingly, the preset score threshold may be a reading accuracy score when the reading accuracy is 100%.

[0171] If the score of the reading speech does not reach the preset score threshold, step S302 is further executed; if the score of the reading speech reaches the preset score threshold, the effective speech duration of the reading speech can be determined as the fluency evaluation result of the reading speech.

[0172] In an optional implementation of the present application, the binary classification model can be obtained by performing scoring training on a pre-built binary classification model using sample speech labeled with scores.

[0173] Furthermore, when there are a large number of reading speech samples to be evaluated, in order to further improve the accuracy and rationality of the speech evaluation, the average value of the fluency evaluation results of the reading speech can be used as the standard value to further rate the fluency of the reading speech. In the embodiment of the present application, since the fluency rating of the reading speech is based on the score of the time dimension, the fluency of the reading speech is inversely proportional to the fluency evaluation result. Based on this, reading speech samples less than or equal to the average value can be rated as excellent, and reading speech samples greater than the average value can be rated as poor.

[0174] It can be understood that the above-mentioned method of rating the fluency of the reading speech is only an optional implementation method provided by the present application for the convenience of understanding. Different methods can also be used to rate the fluency of the reading speech. For example, based on the fluency evaluation results of the reading speech, the fluency rating of the reading speech ranked in the bottom 25% can be determined as poor, the fluency rating of the reading speech ranked in the top 25% can be determined as excellent, and the fluency rating of other reading speech can be determined as good.

[0175] In summary, the speech evaluation method provided in the embodiments of the present application, after obtaining the reading speech, determines the reading accuracy of the reading speech based on the reading speech and the standard text, and evaluates the reading speech based on the reading accuracy and the effective speech duration of the reading speech. This method abandons the traditional method of manual recording and evaluates the reading speech by combining two different dimensions of speech data: reading accuracy and effective speech duration. This method not only improves the efficiency of speech evaluation, but also improves the accuracy of the speech evaluation results.

[0176] Exemplary devices

[0177] Correspondingly, the present application embodiment also provides a voice evaluation device, please refer to Figure 4 , Figure 4 This is a structural diagram of a speech evaluation device provided in another embodiment of the present application.

[0178] like Figure 4 As shown, the speech evaluation device includes:

[0179] The first unit 401 is used to obtain a reading speech generated by the subject reading a standard text and determine the effective speech duration of the reading speech;

[0180] The second unit 402 is configured to obtain a reading accuracy of the reading speech based on the reading speech and the standard text;

[0181] The third unit 403 is configured to generate a fluency evaluation result for the reading speech based on the effective speech duration of the reading speech and the reading accuracy of the reading speech.

[0182] In an optional implementation manner of the present application, the first unit 401 includes:

[0183] The alignment subunit is used to align the standard text with the reading speech and determine the effective speech duration of the reading speech.

[0184] In an optional implementation manner of the present application, the second unit 402 includes:

[0185] an error information acquisition subunit, configured to obtain reading error information in the reading speech according to the reading speech and the standard text; the reading error information includes at least one of phoneme reading error information and text reading error information;

[0186] a first accuracy determination subunit, configured to obtain the reading accuracy of the reading speech based on the reading error information; or to obtain the reading accuracy of the reading speech based on the reading error information and feature information of the reading speech;

[0187] The characteristic information of the reading voice includes at least one of the recognized text of the reading voice and the effective voice duration of the reading voice.

[0188] In an optional implementation manner of the present application, obtaining reading error information in the reading speech according to the reading speech and the standard text includes:

[0189] The reading speech and the standard text are input into a pre-trained error detection model, so that the error detection model performs error detection processing on the reading speech based on the standard text to obtain reading error information in the reading speech.

[0190] In an optional implementation manner of the present application, the second unit 402 includes:

[0191] A speech recognition subunit, configured to perform speech recognition processing on the reading speech to obtain a recognition text corresponding to the reading speech;

[0192] a common text determination subunit, configured to determine a common text sequence between the recognition text and the standard text;

[0193] The second accuracy determination subunit is configured to obtain the reading accuracy of the reading speech according to the common text sequence and the standard text.

[0194] In an optional implementation manner of the present application, obtaining the reading accuracy of the reading speech according to the common text sequence and the standard text includes:

[0195] Obtaining a first score for the reading speech according to a sequence length of the common text sequence;

[0196] Inputting the reading speech and the standard text into a pre-trained error detection model, so that the error detection model performs error detection processing on the reading speech based on the standard text, and obtains text reading error information in the reading speech;

[0197] Obtaining a second score for the reading speech according to the text reading error information;

[0198] Performing linear regression on the first score and the second score to obtain the reading accuracy of the reading speech.

[0199] In an optional embodiment of the present application, after performing speech recognition processing on the reading speech to obtain a recognized text corresponding to the reading speech, the apparatus is further configured to:

[0200] Inputting the reading speech and the recognized text into a pre-trained error detection model, so that the error detection model performs error detection processing on the reading speech based on the recognized text, and obtains text reading error information in the reading speech;

[0201] Based on the text reading error information, the recognized text is subjected to denoising processing to obtain a recognized text after denoising processing, wherein the recognized text after denoising processing includes a correctly read text sequence;

[0202] The determining of a common text sequence between the recognition text and the standard text includes:

[0203] Determine a common text sequence between the recognized text after denoising and the standard text.

[0204] In an optional embodiment of the present application, the error detection model is trained in the following manner:

[0205] Obtaining the first sample speech with standard reading error information;

[0206] The first sample speech is used as a training sample to perform error detection training on a pre-built error detection model.

[0207] In an optional implementation manner of the present application, the third unit 403 includes:

[0208] A penalty duration subunit, configured to generate a penalty duration in a time dimension according to the reading accuracy;

[0209] The superposition subunit is used to superimpose the penalty duration on the effective duration of the reading speech to generate a fluency evaluation result for the reading speech.

[0210] In an optional implementation manner of the present application, before obtaining the reading accuracy of the reading speech based on the reading speech and the standard text, the apparatus is further configured to:

[0211] Inputting the reading speech into a pre-trained binary classification model so that the binary classification model scores the reading speech and determines whether the score of the reading speech reaches a preset scoring threshold;

[0212] If the score of the reading speech does not reach the preset score threshold, performing the step of obtaining the reading accuracy of the reading speech based on the reading speech and the standard text;

[0213] If the score of the reading speech reaches a preset score threshold, a fluency evaluation result for the reading speech is generated based on the effective speech duration of the reading speech.

[0214] In an optional embodiment of the present application, the binary classification model is obtained by:

[0215] Obtaining a second sample speech labeled with a score;

[0216] The second sample speech is used as a training sample to perform scoring training on the pre-built binary classification model.

[0217] The speech evaluation device provided in this embodiment is based on the same concept as the speech evaluation method provided in the aforementioned embodiments of this application. It can execute the speech evaluation method provided in any of the aforementioned embodiments of this application and possesses the corresponding functional modules and beneficial effects of executing the aforementioned speech evaluation method. For technical details not fully described in this embodiment, please refer to the specific processing content of the speech evaluation method provided in the aforementioned embodiments of this application and will not be further elaborated here.

[0218] Exemplary electronic devices

[0219] Another embodiment of the present application also provides an electronic device, please refer to Figure 5 As shown, Figure 5 This is a schematic diagram of the structure of an electronic device provided in another embodiment of the present application, the device comprising:

[0220] Memory 200 and processor 210;

[0221] The memory 200 is connected to the processor 210 and is used to store programs;

[0222] The processor 210 is configured to implement the speech evaluation method disclosed in any of the above embodiments by running the program stored in the memory 200 .

[0223] Specifically, the electronic device may further include: a bus, a communication interface 220 , an input device 230 and an output device 240 .

[0224] The processor 210, the memory 200, the communication interface 220, the input device 230 and the output device 240 are interconnected via a bus.

[0225] A bus may include a pathway that transfers information between components of a computer system.

[0226] Processor 210 can be a general-purpose processor, such as a general-purpose central processing unit (CPU), a microprocessor, or the like, or an application-specific integrated circuit (ASIC), or one or more integrated circuits for controlling the execution of the program of the present invention. Alternatively, it can be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic device, discrete gate or transistor logic device, or discrete hardware components.

[0227] The processor 210 may include a main processor, and may also include a baseband chip, a modem, and the like.

[0228] The memory 200 stores a program for executing the technical solution of the present invention, and may also store an operating system and other key services. Specifically, the program may include program code, which includes computer operating instructions. More specifically, the memory 200 may include read-only memory (ROM), other types of static storage devices that can store static information and instructions, random access memory (RAM), other types of dynamic storage devices that can store information and instructions, disk storage, flash memory, etc.

[0229] The input device 230 may include a device for receiving data and information input by a user, such as a keyboard, a mouse, a camera, a scanner, a light pen, a voice input device, a touch screen, a pedometer, or a gravity sensor.

[0230] Output device 240 may include devices that allow information to be output to a user, such as a display screen, printer, speakers, etc.

[0231] The communication interface 220 may include any device such as a transceiver to communicate with other devices or communication networks, such as Ethernet, Radio Access Network (RAN), Wireless Local Area Network (WLAN), etc.

[0232] The processor 210 executes the program stored in the memory 200 and calls other devices, and can be used to implement each step of any speech evaluation method provided in the above embodiments of the present application.

[0233] Exemplary computer program products and storage media

[0234] In addition to the above-mentioned methods and devices, an embodiment of the present application may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to execute the steps of the speech evaluation method according to various embodiments of the present application described in the above-mentioned "Exemplary Method" section of this specification.

[0235] The computer program product may be written in any combination of one or more programming languages to implement the program code for performing the operations of the embodiments of the present application, including object-oriented programming languages such as Java, C++, and conventional procedural programming languages such as "C" or similar programming languages. The program code may be executed entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0236] In addition, an embodiment of the present application may also be a storage medium having a computer program stored thereon, and the computer program is used by a processor to execute the steps of the speech evaluation method according to various embodiments of the present application described in the "Exemplary Method" section above of this specification, specifically the following steps:

[0237] Step S301, obtaining a reading speech generated by the subject reading a standard text, and determining an effective speech duration of the reading speech;

[0238] Step S302, obtaining the reading accuracy of the reading speech according to the reading speech and the standard text;

[0239] Step S303 : generating a fluency evaluation result for the reading speech based on the effective speech duration of the reading speech and the reading accuracy of the reading speech.

[0240] For the sake of simplicity, the aforementioned method embodiments are described as a series of action combinations. However, those skilled in the art should be aware that this application is not limited by the order of the actions described, because according to this application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily required by this application.

[0241] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similarities between the various embodiments can be referred to in conjunction with each other. For device embodiments, since they are generally similar to method embodiments, their description is relatively simple, and for relevant details, reference can be made to the description of the method embodiments.

[0242] The steps in the methods of each embodiment of the present application can be adjusted in sequence, merged, and deleted according to actual needs, and the technical features recorded in each embodiment can be replaced or combined.

[0243] The modules and sub-modules in the devices and terminals of the various embodiments of the present application can be merged, divided, and deleted according to actual needs.

[0244] In the several embodiments provided in this application, it should be understood that the disclosed terminals, devices, and methods can be implemented in other ways. For example, the terminal embodiments described above are merely illustrative. For example, the division of modules or submodules is merely a logical function division. In actual implementation, there may be other division methods, such as multiple submodules or modules can be combined or integrated into another module, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or module, which can be electrical, mechanical or other forms.

[0245] The modules or submodules described as separate components may or may not be physically separate, and the components of the modules or submodules may or may not be physical modules or submodules, that is, they may be located in one place or distributed across multiple network modules or submodules. Some or all of the modules or submodules may be selected to achieve the purpose of this embodiment according to actual needs.

[0246] In addition, each functional module or submodule in each embodiment of the present application may be integrated into a processing module, or each module or submodule may exist physically separately, or two or more modules or submodules may be integrated into a single module. The above-mentioned integrated modules or submodules may be implemented in the form of hardware or software functional modules or submodules.

[0247] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0248] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, software units executed by a processor, or a combination of the two. The software units may be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0249] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0250] The above description of the disclosed embodiments will enable those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is to be construed in the widest manner consistent with the principles and novel features disclosed herein.

Claims

1. A speech evaluation method, characterized in that: include: Obtaining reading speech generated by the subject reading a standard text, and determining the effective speech duration of the reading speech; wherein the standard text includes a question book for a rapid naming task; Inputting the reading speech into a pre-trained binary classification model so that the binary classification model scores the reading speech and determines whether the score of the reading speech reaches a preset scoring threshold; If the score of the reading speech does not reach a preset score threshold, obtaining the reading accuracy of the reading speech based on the reading speech and the standard text; and generating a fluency evaluation result for the reading speech based on the effective speech duration of the reading speech and the reading accuracy of the reading speech; If the score of the reading speech reaches a preset score threshold, a fluency evaluation result for the reading speech is generated based on the effective speech duration of the reading speech.

2. The method according to claim 1, characterized in that Determining the effective voice duration of the reading voice includes: The standard text and the reading speech are aligned to determine the effective speech duration of the reading speech.

3. The method according to claim 1, characterized in that The obtaining of the reading accuracy of the reading speech according to the reading speech and the standard text includes: Obtaining reading error information in the reading speech according to the reading speech and the standard text; the reading error information includes at least one of phoneme reading error information and text reading error information; Obtaining the reading accuracy of the reading speech according to the reading error information; or obtaining the reading accuracy of the reading speech according to the reading error information and feature information of the reading speech; The characteristic information of the reading voice includes at least one of the recognized text of the reading voice and the effective voice duration of the reading voice.

4. The method according to claim 3, characterized in that The obtaining, based on the reading voice and the standard text, reading error information in the reading voice includes: The reading speech and the standard text are input into a pre-trained error detection model, so that the error detection model performs error detection processing on the reading speech based on the standard text to obtain reading error information in the reading speech.

5. The method according to claim 1, wherein The obtaining of the reading accuracy of the reading speech according to the reading speech and the standard text includes: Performing speech recognition processing on the reading speech to obtain a recognition text corresponding to the reading speech; Determining a common text sequence between the recognition text and the standard text; The reading accuracy of the reading speech is obtained according to the common text sequence and the standard text.

6. The method according to claim 5, characterized in that The obtaining of the reading accuracy of the reading speech according to the common text sequence and the standard text includes: Obtaining a first score for the reading speech according to a sequence length of the common text sequence; Inputting the reading speech and the standard text into a pre-trained error detection model, so that the error detection model performs error detection processing on the reading speech based on the standard text, and obtains text reading error information in the reading speech; Obtaining a second score for the reading speech according to the text reading error information; Performing linear regression on the first score and the second score to obtain the reading accuracy of the reading speech.

7. The method according to claim 5, characterized in that After performing speech recognition processing on the reading speech to obtain a recognized text corresponding to the reading speech, the method further includes: Inputting the reading speech and the recognized text into a pre-trained error detection model, so that the error detection model performs error detection processing on the reading speech based on the recognized text, and obtains text reading error information in the reading speech; Based on the text reading error information, the recognized text is subjected to denoising processing to obtain a recognized text after denoising processing, wherein the recognized text after denoising processing includes a correctly read text sequence; The determining of a common text sequence between the recognition text and the standard text includes: Determine a common text sequence between the recognized text after denoising and the standard text.

8. The method according to any one of claims 4, 6 and 7, characterized in that: The error detection model is trained in the following way: Obtaining the first sample speech with standard reading error information; The first sample speech is used as a training sample to perform error detection training on a pre-built error detection model.

9. The method according to claim 1, characterized in that Generating a fluency evaluation result for the reading speech based on the effective speech duration of the reading speech and the reading accuracy of the reading speech includes: Generating a penalty duration in a time dimension according to the reading accuracy; The penalty duration is added to the effective duration of the reading speech to generate a fluency evaluation result for the reading speech.

10. The method according to claim 1, characterized in that The two-class model is obtained in the following way: Obtaining a second sample speech labeled with a score; The second sample speech is used as a training sample to perform scoring training on the pre-built binary classification model.

11. A speech evaluation device, characterized in that: include: The first unit is used to obtain the reading speech generated by the test subject when reading the standard text and determine the effective speech duration of the reading speech; wherein the standard text includes a question book for the rapid naming task; A scoring unit, configured to input the reading speech into a pre-trained binary classification model so that the binary classification model scores the reading speech and determines whether the score of the reading speech reaches a preset scoring threshold; The second unit is configured to obtain a reading accuracy of the reading speech based on the reading speech and the standard text if the score of the reading speech does not reach a preset score threshold; A third unit is configured to generate a fluency evaluation result for the reading speech based on the effective speech duration of the reading speech and the reading accuracy of the reading speech; The fourth unit is configured to generate a fluency evaluation result for the reading speech based on the effective speech duration of the reading speech if the score of the reading speech reaches a preset score threshold.

12. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the speech evaluation method according to any one of claims 1 to 10 by running instructions in the memory.

13. A computer-readable storage medium, characterized in that The storage medium stores a computer program, and when the computer program is executed by the processor, the speech evaluation method according to any one of claims 1 to 10 is executed.

Images