Intelligent evaluation method and system for oral English

Abstract

The invention relates to the technical field of speech recognition, and particularly discloses an intelligent evaluation method and system for spoken English, and the method comprises the following steps: S1, extracting spoken language features based on spoken language test speech, determining target features based on the spoken language features, and constructing a feature triangle based on the target features; s2, setting an area grade in the area range, and determining the area grade of the feature triangle according to the area of the feature triangle; the target triangle with the largest area is used as a reference triangle, and the target triangles are grouped according to the area relation between the target triangles and the reference triangle; s3, obtaining a target triangle sequence, and determining representative triangles in the target triangle sequence; and constructing a standard triangle based on the standard speech, determining the similarity between the representative triangle and the standard triangle, and determining a spoken language ability evaluation result based on the similarity. The oral English evaluation efficiency can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of speech recognition, and in particular to an intelligent evaluation method and system for spoken English. Background Art

[0002] Spoken English is an oral expression form that uses English as the medium of communication. It is mainly used for instant communication in daily life, work and study. It is a form of language practice that emphasizes pronunciation, fluency and naturalness of expression.

[0003] At present, the oral English proficiency assessment is mainly conducted through interviews, where examiners determine the oral proficiency of candidates. However, the interviews are scored by examiners, whose personal preferences, experience, and status on the day may affect the scoring, resulting in a lack of consistency and objectivity in the assessment results. For example, different examiners may have significant differences in their evaluation of the same expression. In addition, interview assessments require face-to-face or online interaction between examiners and candidates, which is time- and manpower-intensive, and is difficult to meet the needs of efficiency and economy, especially in large-scale testing. Summary of the invention

[0004] The purpose of the present invention is to provide an intelligent evaluation method and system for spoken English to solve the following technical problems: The examiner's personal preference, experience and state on the day may affect the scoring, resulting in a lack of consistency and objectivity in the evaluation results. For example, different examiners may have significantly different evaluations of the same expression. In addition, interview evaluation requires the examiner to interact with the candidate face-to-face or online. This method has high requirements on time and manpower, especially in large-scale testing, and it is difficult to meet the needs of efficiency and economy.

[0005] The purpose of the present invention can be achieved through the following technical solutions: An intelligent evaluation method for spoken English, comprising the following steps: S1: obtaining the oral test speech of the examinee, extracting oral features based on the oral test speech, wherein the oral features include fundamental frequency, formant and short-time energy, preprocessing the oral features to obtain target features, and using the target features as the lengths of the three sides of the triangle to construct a feature triangle; S2: Set area range [ / 4, ], setting a plurality of area levels from low to high at preset area intervals within the area range, and determining the area level of the characteristic triangle according to the area of ​​the characteristic triangle; Mark the characteristic triangles in the same area level as target triangles, take the target triangle with the largest area as the reference triangle D, determine the area of ​​the overlapping part of the target triangle and the reference triangle D, determine the ratio C of the area of ​​the overlapping part to the target triangle, group the target triangles based on the ratio C, if the difference of the ratio C corresponding to any two target triangles in the same group is less than a preset threshold, the reference triangle D belongs to the first group, and the first group is the group where the target triangle with the largest area except the reference triangle D is located; S3: sorting the target triangles in the same group in descending order according to their area sizes to obtain a target triangle sorting, determining a target triangle corresponding to a median in the target triangle sorting, and using the target triangle as a representative triangle; Obtain a preset standard voice, determine the spoken features of the standard voice and construct a feature triangle, use it as the standard triangle, determine the similarity P between the representative triangle and the standard triangle, and determine the oral ability assessment result of the examinee based on the similarity P.

[0006] As a further solution of the present invention: in step S3, the process of determining the similarity P between the representative triangle and the standard triangle specifically includes: Calculate the area difference ΔS=S DB -S BZ , S DB and S BZ Respectively represent the areas of the representative triangle and the standard triangle; Determining an angle ratio, wherein the angle ratio is the ratio of the angles between two identical sides of the representative triangle and the standard triangle; Calculate the similarity P = ε * ΔS + A, A represents the sum of the angle ratios, and ε is the preset correction coefficient.

[0007] As a further solution of the present invention: in step S3, the process of determining the test-taker's oral proficiency assessment result specifically includes: Set judgment thresholds P1 and P2, and P1<P2; When 0<P≤P1, the similarity is low, and the examinee’s oral ability is judged to be poor; When P1<P≤P2, the similarity is average, and the examinee’s oral ability is judged to be average; When P2<P, the similarity is high, and the examinee’s oral ability is judged to be good. The oral ability assessment results of candidates in the same group are the same and are all assessment results corresponding to the similarity P.

[0008] As a further solution of the present invention: in the step S3, during the process of determining the target triangle sorting, target triangles with the same area are in the same position in the target triangle sorting.

[0009] As a further solution of the present invention: in the step S3, when the median in the target triangle sorting is not an integer, the two target triangles adjacent to the median are regarded as pending triangles, the similarity between the pending triangles and the standard triangle is determined, and the mean is calculated as the similarity P.

[0010] As a further solution of the present invention: in the step S2, in the process of determining the area of ​​the characteristic triangle, the area of ​​the characteristic triangle is calculated with the side corresponding to the same target feature as the base.

[0011] As a further solution of the present invention: in the step S1, the process of preprocessing the spoken language features specifically includes: The spoken language features are standardized to eliminate the dimension, and the standardized spoken language features are mapped to a preset value range [1,2] to obtain the target features.

[0012] An intelligent evaluation system for spoken English, comprising: Conversion module: obtaining the oral test speech of the examinee, extracting oral features based on the oral test speech, wherein the oral features include fundamental frequency, formant and short-time energy, preprocessing the oral features to obtain target features, using the target features as the lengths of the three sides of the triangle, and constructing a feature triangle; Grouping module: Set area range [ / 4, ], setting a plurality of area levels from low to high at preset area intervals within the area range, and determining the area level of the characteristic triangle according to the area of ​​the characteristic triangle; Mark the characteristic triangles in the same area level as target triangles, take the target triangle with the largest area as the reference triangle D, determine the area of ​​the overlapping part of the target triangle and the reference triangle D, determine the ratio C of the area of ​​the overlapping part to the target triangle, group the target triangles based on the ratio C, if the difference of the ratio C corresponding to any two target triangles in the same group is less than a preset threshold, the reference triangle D belongs to the first group, and the first group is the group where the target triangle with the largest area except the reference triangle D is located; Evaluation module: sorting the target triangles in the same group in descending order according to their area sizes to obtain a target triangle sorting, determining a target triangle corresponding to the median in the target triangle sorting, and taking it as a representative triangle; Obtain a preset standard voice, determine the spoken features of the standard voice and construct a feature triangle, use it as the standard triangle, determine the similarity P between the representative triangle and the standard triangle, and determine the oral ability assessment result of the examinee based on the similarity P.

[0013] The beneficial effects of the present invention are as follows: in this scheme, through preprocessing, the data dimension differences between different examinees are eliminated, so that the features are comparable, the standardized target features make the data of all examinees at the same scale, which is convenient for batch processing, and the complex speech signals are converted into simple geometric models (feature triangles) to represent and compare the oral features of different examinees in an intuitive way, which is convenient for subsequent calculation and comparison, thereby improving the processing efficiency, simplifying the multi-dimensional speech features into geometric shapes, reducing the calculation complexity, and the unified feature representation form enables the system to process the data of a large number of examinees at the same time, thereby improving the processing speed; examinees with similar features are classified together according to the area and overlap, ensuring that feature triangles with large differences in area are not classified into the same group, through This method avoids the time consumed by detailed analysis, and then uses the overlap ratio C to classify triangles with similar areas in more detail. After grouping, the oral characteristics of candidates in the same group are more similar, and the evaluation results are more representative. The data of a large number of candidates are summarized into a small number of groups, which reduces the scale of data for subsequent processing. Through grouping, only the representative data of each group needs to be calculated instead of each candidate, thereby improving the overall processing efficiency; by evaluating the representative triangles, the oral ability level of the entire group is inferred, avoiding repeated similarity calculations for each candidate. Only one similarity calculation is needed to obtain the evaluation result of the entire group, which greatly reduces the number of similarity calculations, thereby reducing processing time and speeding up the overall evaluation process. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The present invention will be further described below in conjunction with the accompanying drawings.

[0015] Figure 1 It is a flow chart of an intelligent evaluation method for spoken English of the present invention. DETAILED DESCRIPTION

[0016] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0017] See also Figure 1 As shown, the present invention is an intelligent evaluation method for spoken English, comprising the following steps: S1: obtaining the oral test speech of the examinee, extracting oral features based on the oral test speech, wherein the oral features include fundamental frequency, formant and short-time energy, preprocessing the oral features to obtain target features, and using the target features as the lengths of the three sides of the triangle to construct a feature triangle; S2: Set area range [ / 4, ], setting a plurality of area levels from low to high at preset area intervals within the area range, and determining the area level of the characteristic triangle according to the area of ​​the characteristic triangle; Mark the characteristic triangles in the same area level as target triangles, take the target triangle with the largest area as the reference triangle D, determine the area of ​​the overlapping part of the target triangle and the reference triangle D, determine the ratio C of the area of ​​the overlapping part to the target triangle, group the target triangles based on the ratio C, if the difference of the ratio C corresponding to any two target triangles in the same group is less than a preset threshold, the reference triangle D belongs to the first group, and the first group is the group where the target triangle with the largest area except the reference triangle D is located; S3: sorting the target triangles in the same group in descending order according to their area sizes to obtain a target triangle sorting, determining a target triangle corresponding to a median in the target triangle sorting, and using the target triangle as a representative triangle; Obtain a preset standard voice, determine the spoken features of the standard voice and construct a feature triangle, use it as the standard triangle, determine the similarity P between the representative triangle and the standard triangle, and determine the oral ability assessment result of the examinee based on the similarity P.

[0018] It should be noted that, through preprocessing, the differences in data dimensions between different examinees are eliminated, making the features comparable. The standardized target features make the data of all examinees on the same scale, which is convenient for batch processing. The complex speech signals are converted into simple geometric models (feature triangles) to represent and compare the oral features of different examinees in an intuitive way, which is convenient for subsequent calculations and comparisons, thereby improving processing efficiency. The multi-dimensional speech features are simplified into geometric shapes, which reduces the calculation complexity. The unified feature representation allows the system to process the data of a large number of examinees at the same time, improving the processing speed. According to the area and overlap, examinees with similar features are classified together to ensure that feature triangles with large differences in area are not classified in the same group. In this way, To avoid time consumption for detailed analysis, the overlap ratio C is then used to perform a more detailed classification of triangles with similar areas. After grouping, the oral characteristics of candidates in the same group are more similar, and the evaluation results are more representative. The data of a large number of candidates are summarized into a small number of groups, which reduces the scale of data for subsequent processing. Through grouping, only the representative data of each group needs to be calculated instead of each candidate, thereby improving the overall processing efficiency; by evaluating the representative triangles, the oral ability level of the entire group is inferred, avoiding repeated similarity calculations for each candidate. Only one similarity calculation is needed to obtain the evaluation result of the entire group, which greatly reduces the number of similarity calculations, thereby reducing processing time and speeding up the overall evaluation process.

[0019] In another preferred embodiment of the present invention, in step S3, the process of determining the similarity P between the representative triangle and the standard triangle specifically includes: Calculate the area difference ΔS=S DB -S BZ , S DB and S BZ Respectively represent the areas of the representative triangle and the standard triangle; Determining an angle ratio, wherein the angle ratio is the ratio of the angles between two identical sides of the representative triangle and the standard triangle; Calculate the similarity P = ε * ΔS + A, A represents the sum of the angle ratios, and ε is the preset correction coefficient.

[0020] It is worth noting that by combining the area difference and the angle ratio, the geometric similarity between the representative triangle and the standard triangle is comprehensively evaluated, making the similarity calculation more precise and reflecting the differences between the examinee's oral characteristics and the standard oral characteristics; among them, the process of obtaining the angle ratio, such as the two sides JPB and GZFB of the target feature corresponding to the fundamental frequency and the resonance peak, determines the angle DBJ between the sides JPB and GZFB of the representative triangle and the angle BZJ between the sides JPB and GZFB of the standard triangle, then the angle ratio is equal to DBJ / BZJ.

[0021] In another preferred embodiment of the present invention, in step S3, the process of determining the test-taker's oral proficiency assessment result specifically includes: Set judgment thresholds P1 and P2, and P1<P2; When 0<P≤P1, the similarity is low, and the examinee’s oral ability is judged to be poor; When P1<P≤P2, the similarity is average, and the examinee’s oral ability is judged to be average; When P2<P, the similarity is high, and the examinee’s oral ability is judged to be good; The oral ability assessment results of candidates in the same group are the same and are all assessment results corresponding to the similarity P.

[0022] It is understandable that since the oral characteristics of candidates in the same group are similar, giving the same evaluation results can ensure the fairness and consistency of the evaluation. Therefore, by calculating the similarity of the representative triangles in the group once and applying the result to the entire group, it is possible to avoid calculating the similarity for each candidate one by one, which significantly reduces the calculation workload and shortens the time required for evaluation, making it suitable for large-scale oral tests.

[0023] In another preferred embodiment of the present invention, in the step S3, during the process of determining the target triangle sorting, target triangles with the same area are in the same position in the target triangle sorting.

[0024] In another preferred embodiment of the present invention, in the step S3, when the median in the target triangle sorting is not an integer, two target triangles adjacent to the median are regarded as pending triangles, the similarity between the pending triangles and the standard triangle is determined, and the mean is calculated as the similarity P.

[0025] It should be noted that the similarity between the undetermined triangle and the standard triangle is calculated in the same way as the similarity between the representative triangle and the standard triangle.

[0026] In another preferred embodiment of the present invention, in the step S2, in the process of determining the area of ​​the characteristic triangle, the area of ​​the characteristic triangle is calculated with the side corresponding to the same target feature as the base.

[0027] In another preferred embodiment of the present invention, in step S1, the process of preprocessing the spoken language features specifically includes: The spoken language features are standardized to eliminate the dimension, and the standardized spoken language features are mapped to a preset value range [1,2] to obtain the target features.

[0028] It is worth noting that An intelligent evaluation system for spoken English, comprising: Conversion module: obtaining the oral test speech of the examinee, extracting oral features based on the oral test speech, wherein the oral features include fundamental frequency, formant and short-time energy, preprocessing the oral features to obtain target features, using the target features as the lengths of the three sides of the triangle, and constructing a feature triangle; Grouping module: Set area range [ / 4, ], setting a plurality of area levels from low to high at preset area intervals within the area range, and determining the area level of the characteristic triangle according to the area of ​​the characteristic triangle; Mark the characteristic triangles in the same area level as target triangles, take the target triangle with the largest area as the reference triangle D, determine the area of ​​the overlapping part of the target triangle and the reference triangle D, determine the ratio C of the area of ​​the overlapping part to the target triangle, group the target triangles based on the ratio C, if the difference of the ratio C corresponding to any two target triangles in the same group is less than a preset threshold, the reference triangle D belongs to the first group, and the first group is the group where the target triangle with the largest area except the reference triangle D is located; Evaluation module: sorting the target triangles in the same group in descending order according to their area sizes to obtain a target triangle sorting, determining a target triangle corresponding to the median in the target triangle sorting, and taking it as a representative triangle; Obtain a preset standard voice, determine the spoken features of the standard voice and construct a feature triangle, use it as the standard triangle, determine the similarity P between the representative triangle and the standard triangle, and determine the oral ability assessment result of the examinee based on the similarity P.

[0029] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. An intelligent evaluation method for spoken English, characterized in that: The following steps are involved: S1: obtaining the oral test speech of the examinee, extracting oral features based on the oral test speech, wherein the oral features include fundamental frequency, formant and short-time energy, preprocessing the oral features to obtain target features, and using the target features as the lengths of the three sides of the triangle to construct a feature triangle; S2: Set area range [ / 4, ], setting a plurality of area levels from low to high at preset area intervals within the area range, and determining the area level of the characteristic triangle according to the area of ​​the characteristic triangle; Mark the characteristic triangles in the same area level as target triangles, take the target triangle with the largest area as the reference triangle D, determine the area of ​​the overlapping part of the target triangle and the reference triangle D, determine the ratio C of the area of ​​the overlapping part to the target triangle, group the target triangles based on the ratio C, if the difference of the ratio C corresponding to any two target triangles in the same group is less than a preset threshold, the reference triangle D belongs to the first group, and the first group is the group where the target triangle with the largest area except the reference triangle D is located; S3: sorting the target triangles in the same group in descending order according to their area sizes to obtain a target triangle sorting, determining a target triangle corresponding to a median in the target triangle sorting, and using the target triangle as a representative triangle; Obtain a preset standard voice, determine the spoken features of the standard voice and construct a feature triangle, use it as the standard triangle, determine the similarity P between the representative triangle and the standard triangle, and determine the oral ability assessment result of the examinee based on the similarity P.

2. The intelligent evaluation method for spoken English according to claim 1, characterized in that: In step S3, the process of determining the similarity P between the representative triangle and the standard triangle specifically includes: Calculate the area difference ΔS=S DB -S BZ , S DB and S BZ Respectively represent the areas of the representative triangle and the standard triangle; Determining an angle ratio, wherein the angle ratio is the ratio of the angles between two identical sides of the representative triangle and the standard triangle; Calculate the similarity P = ε * ΔS + A, A represents the sum of the angle ratios, and ε is the preset correction coefficient.

3. The intelligent evaluation method for spoken English according to claim 1, characterized in that: In step S3, the process of determining the test taker's oral proficiency assessment result specifically includes: Set judgment thresholds P1 and P2, and P1<P2; When 0<P≤P1, the similarity is low, and the examinee’s oral ability is judged to be poor; When P1<P≤P2, the similarity is average, and the examinee’s oral ability is judged to be average; When P2<P, the similarity is high, and the examinee’s oral ability is judged to be good; The oral ability assessment results of candidates in the same group are the same and are all assessment results corresponding to the similarity P.

4. The intelligent evaluation method for spoken English according to claim 1, characterized in that: In the step S3, during the process of determining the target triangle sorting, target triangles with the same area are in the same position in the target triangle sorting.

5. The intelligent evaluation method for spoken English according to claim 1, characterized in that: In the step S3, when the median in the target triangle sorting is not an integer, two target triangles adjacent to the median are regarded as pending triangles, the similarity between the pending triangles and the standard triangle is determined, and the mean is calculated as the similarity P.

6. The intelligent evaluation method for spoken English according to claim 1, characterized in that: In the step S2, in the process of determining the area of ​​the characteristic triangle, the area of ​​the characteristic triangle is calculated with the side corresponding to the same target feature as the base.

7. The intelligent evaluation method for spoken English according to claim 1, characterized in that: In the step S1, the process of preprocessing the spoken language features specifically includes: The spoken language features are standardized to eliminate the dimension, and the standardized spoken language features are mapped to a preset value range [1,2] to obtain the target features.

8. An intelligent evaluation system for spoken English, characterized in that: include: Conversion module: obtaining the oral test speech of the examinee, extracting oral features based on the oral test speech, wherein the oral features include fundamental frequency, formant and short-time energy, preprocessing the oral features to obtain target features, using the target features as the lengths of the three sides of the triangle, and constructing a feature triangle; Grouping module: Set area range [ / 4, ], setting a plurality of area levels from low to high at preset area intervals within the area range, and determining the area level of the characteristic triangle according to the area of ​​the characteristic triangle; Mark the characteristic triangles in the same area level as target triangles, take the target triangle with the largest area as the reference triangle D, determine the area of ​​the overlapping part of the target triangle and the reference triangle D, determine the ratio C of the area of ​​the overlapping part to the target triangle, group the target triangles based on the ratio C, if the difference of the ratio C corresponding to any two target triangles in the same group is less than a preset threshold, the reference triangle D belongs to the first group, and the first group is the group where the target triangle with the largest area except the reference triangle D is located; Evaluation module: sorting the target triangles in the same group in descending order according to their area sizes to obtain a target triangle sorting, determining a target triangle corresponding to the median in the target triangle sorting, and taking it as a representative triangle; Obtain a preset standard voice, determine the spoken features of the standard voice and construct a feature triangle, use it as the standard triangle, determine the similarity P between the representative triangle and the standard triangle, and determine the oral ability assessment result of the examinee based on the similarity P.

Images