Audio information quality measurement method, device and electronic equipment

Abstract

The present application provides a method, device and electronic device for measuring the quality of audio information. After obtaining the text to be processed corresponding to the audio to be processed, entity extraction is performed on the text to be processed to obtain multiple words belonging to the entity. Based on the field to which each word belongs and the type of each word, the score of each word is obtained according to a preset rule, and the rule score of the text to be processed is obtained. In addition, clustering is performed on the multiple words, and the semantic score of the text to be processed is obtained based on the clustering result. According to the rule score and the semantic score, the quality measurement score of the text to be processed is obtained. In this scheme, the quality evaluation of the text is performed based on the field and type of the words, as well as the distribution of the words represented by the clustering results, so that the measurement based on the substantive content of the text can be achieved, and the measurement result that can represent the quality of the substantive content information can be obtained.

Description

Technical Field

[0001] The present application relates to the technical field of natural language processing, and in particular to an audio information quality measurement method, device and electronic device. Background Art

[0002] On large audio sharing platforms, users can freely upload and share original albums and use the platform to search and listen according to their own needs. Due to the uncertainty of user quality, albums naturally present a mixed quality. Therefore, a priori measurement of the quality of albums will help the platform provide better recommendation services. At the same time, pre-screening based on the quality scoring system can also ensure the basic quality of albums at the overall level.

[0003] However, the existing technology does not have a scientific and reasonable quality measurement solution in this field. The measurement process is often based on the sound quality and sound effects of the audio itself, and is not measured from the perspective of the content information quality of the audio work for the audience. There is a lack of measurement indicators that are of substantial help. Summary of the invention

[0004] The purpose of the present application includes, for example, providing a method, device and electronic device for measuring the quality of audio information, which can obtain a measurement result that can characterize the quality of the substantial content information based on the measurement of the substantial content of the text converted from audio.

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

[0006] In a first aspect, the present application provides a method for measuring audio information quality, the method comprising:

[0007] Recognize the audio to be processed and obtain the corresponding text to be processed;

[0008] Performing entity extraction on the text to be processed to obtain multiple words belonging to the entity;

[0009] Based on the fields to which each of the words belongs and the types of each of the words, the scores of each of the words are obtained according to preset rules, and the rule scores of the text to be processed are calculated based on the scores of the multiple words;

[0010] Performing clustering processing on the multiple words, and obtaining a semantic score of the text to be processed based on the clustering result;

[0011] A quality measurement score of the text to be processed is obtained according to the rule score and the semantic score.

[0012] In an optional implementation manner, the text to be processed includes a plurality of sentences, and before the step of performing entity extraction on the text to be processed, the method further includes:

[0013] When the number of sentences contained in the to-be-processed text exceeds a set number, screening out target sentences whose length is less than a preset length from among the multiple sentences;

[0014] Filtering the screened target sentences from the text to be processed;

[0015] If the number of sentences contained in the filtered text to be processed exceeds the set number, the text to be processed is truncated to retain the text to be processed with the set number of sentences.

[0016] In an optional implementation manner, the step of obtaining the score of each word according to a preset rule based on the field to which each word belongs and the type of each word includes:

[0017] For each of the words, if the field to which the word belongs is a first preset field, the score of the word is set to a score corresponding to the first preset field;

[0018] If the field to which the word belongs does not belong to the first preset field, obtaining the type of the word;

[0019] According to the type of the word, a score corresponding to the type is set for the word.

[0020] In an optional implementation manner, the step of setting a score corresponding to the type for the word according to the type of the word includes:

[0021] Detecting whether the length of the word is greater than or equal to a set length, and if so, setting the score of the word to a first score corresponding to its type;

[0022] If the length of the word is less than the set length, the score of the word is set to a second score corresponding to its type, and the second score is greater than the first score.

[0023] In an optional implementation, the step of clustering the multiple words includes:

[0024] Obtain word vectors corresponding to each of the words;

[0025] Calculate the cosine similarity between every two word vectors;

[0026] Multiple word vectors are clustered according to the cosine similarity between every two word vectors.

[0027] In an optional implementation, the step of obtaining a word vector corresponding to each of the words includes:

[0028] For each of the words, a preset vector library is queried. If the word vector corresponding to the word is not found, the word is re-imported into the entity extraction tool for entity extraction, and a query is performed based on the words corresponding to the re-extracted entity until the corresponding word vector is obtained.

[0029] In an optional implementation, the step of obtaining the semantic score of the text to be processed based on the clustering result includes:

[0030] Based on the multiple clusters obtained after clustering, the intra-cluster distance of each cluster and the inter-cluster distance between multiple clusters are calculated;

[0031] The semantic score of the text to be processed is calculated based on the obtained intra-cluster distance and inter-cluster distance.

[0032] In an optional implementation, the step of obtaining a quality measurement score of the text to be processed according to the rule score and the semantic score includes:

[0033] When the domain to which the words contained in the text to be processed belong is the second preset domain, the smaller value between the rule score and the semantic score is used as the quality measurement score, or the minimum value among the rule score, the semantic score and the set score is used as the quality measurement score;

[0034] When the field to which the words contained in the text to be processed belong does not belong to the second preset field, calculating the difference between the rule score and the semantic score;

[0035] If the difference falls within a preset range, the average of the rule score and the semantic score is used as the quality measurement score; if the difference does not fall within the preset range, the rule score is used as the quality measurement score.

[0036] In a second aspect, the present application provides an audio information quality measurement device, the device comprising:

[0037] A recognition module is used to recognize the audio to be processed and obtain the corresponding text to be processed;

[0038] An extraction module, used to perform entity extraction on the text to be processed to obtain multiple words belonging to the entity;

[0039] A calculation module, used to obtain the score of each word according to a preset rule based on the field to which each word belongs and the type of each word, and to calculate the rule score of the text to be processed according to the scores of multiple words;

[0040] A clustering module, used for clustering the multiple words and obtaining a semantic score of the text to be processed based on the clustering result;

[0041] The obtaining module is used to obtain the quality measurement score of the text to be processed according to the rule score and the semantic score.

[0042] In a third aspect, the present application provides an electronic device comprising one or more storage media and one or more processors communicating with the storage media, wherein the one or more storage media store machine-executable instructions executable by the processor, and when the electronic device is running, the processor executes the machine-executable instructions to perform the method steps described in any one of the aforementioned embodiments.

[0043] The beneficial effects of the embodiments of the present application include, for example:

[0044] The present application provides a method, device and electronic device for measuring the quality of audio information. After obtaining the text to be processed corresponding to the audio to be processed, entity extraction is performed on the text to be processed to obtain multiple words belonging to the entity. Based on the field to which each word belongs and the type of each word, the score of each word is obtained according to a preset rule, and the rule score of the text to be processed is obtained. In addition, clustering is performed on the multiple words, and the semantic score of the text to be processed is obtained based on the clustering result. According to the rule score and the semantic score, the quality measurement score of the text to be processed is obtained. In this scheme, the quality evaluation of the text is performed based on the field and type of the words, as well as the distribution of the words represented by the clustering results, so that the measurement based on the substantive content of the text can be achieved, and the measurement result that can represent the quality of the substantive content information can be obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0046] Figure 1 A flowchart of the audio information quality measurement method provided in an embodiment of the present application;

[0047] Figure 2 A flowchart of a filtering method in the audio information quality measurement method provided in an embodiment of the present application;

[0048] Figure 3 for Figure 1 Flow chart of sub-steps included in step S103;

[0049] Figure 4 for Figure 3 Flow chart of sub-steps included in step S1033;

[0050] Figure 5 for Figure 1 Flow chart of sub-steps included in step S104;

[0051] Figure 6 for Figure 1 Another flowchart of the sub-steps included in step S104;

[0052] Figure 7 for Figure 1 Flow chart of sub-steps included in step S105;

[0053] Figure 8 A structural block diagram of an electronic device provided in an embodiment of the present application;

[0054] Fig. 9 This is a functional module block diagram of the audio information quality measurement device provided in an embodiment of the present application.

[0055] Icons: 110 - storage medium; 120 - processor; 130 - audio information quality measurement device; 131 - identification module; 132 - extraction module; 133 - calculation module; 134 - clustering module; 135 - acquisition module; 140 - communication interface. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations.

[0057] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application for which protection is sought, but merely represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0058] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0059] In addition, the terms “first”, “second”, etc., if used, are merely used to distinguish between the descriptions and should not be understood as indicating or implying relative importance.

[0060] See also Figure 1, is a flow chart of the audio information quality measurement method provided in the embodiment of the present application. The method steps defined in the process related to the audio information quality measurement method can be implemented by an electronic device with text analysis and processing functions. Figure 1 The specific process shown is described in detail.

[0061] S101, recognizing the audio to be processed to obtain the corresponding text to be processed.

[0062] S102, performing entity extraction on the text to be processed to obtain a plurality of words belonging to the entity.

[0063] S103, based on the field to which each of the words belongs and the type of each of the words, the score of each of the words is obtained according to a preset rule, and the rule score of the text to be processed is calculated based on the scores of multiple words.

[0064] S104, clustering the multiple words, and obtaining a semantic score of the text to be processed based on the clustering result.

[0065] S105: Obtain a quality measurement score of the text to be processed according to the rule score and the semantic score.

[0066] The audio information quality measurement method provided in this embodiment can be applied to an audio sharing platform, and the audio to be processed can be the audio uploaded by the user to the audio sharing platform, for example, it can be literary audio, financial audio, live audio, etc. The audio to be processed is recognized and processed and converted into the corresponding text to be processed. The recognition processing method can adopt any commonly used audio-to-text processing method, which is not described in detail in this embodiment.

[0067] The text obtained based on audio conversion contains entities and non-entities. Entities refer to words in the text that can objectively represent specific things, such as names of people, places, proper nouns, names of organizations, names of objects, etc. Non-entities can be words such as conjunctions, auxiliary words, modal particles and prepositions that do not have complete lexical meanings and do not express real meanings.

[0068] Non-entity words in the text are not actually helpful for measuring the quality of text information, so in this embodiment, only entities in the text to be processed are extracted to obtain multiple words belonging to the entity. Optionally, entity extraction tools can be used to extract entities.

[0069] Since audios in different fields may not be the same in terms of professionalism, the more professional the audio is, the more helpful it is to the audience, which can be understood as higher quality. Therefore, in this embodiment, the "depth" dimension is adopted, that is, the density of concepts discussed in the audio content in the main field involved. The size of the measurement result on the "depth" dimension is positively correlated with the professionalism of the audio content information. Therefore, audio works with higher depth result scores will intuitively provide higher information quality to the audience.

[0070] The "depth" dimension is often reflected in the field to which the text belongs and the corresponding types of words, or the distribution of words. For example, in fields with high professional requirements such as finance, law, and medicine, if the text (that is, audio) belongs to these fields, it can be understood that the text has a high level of professionalism, that is, the information quality of the text is high. In addition, similarly, for each word contained in the text, the type of the word can also comprehensively reflect the information quality of the text to a certain extent. The types of words can include names of people, places, institutions, works, proper nouns, etc.

[0071] Therefore, in this embodiment, the score of each word can be determined according to a preset rule based on the field to which each word belongs and the type of each word in the text. And based on the sum of the scores of all words contained in the text, the rule score of the text to be processed can be calculated.

[0072] The above method measures the text from the perspective of the field to which the words in the text belong and the type of words. In addition, the text can also be measured from the distribution of the words contained in the text. For example, if the distribution of words of the same category in the text is more concentrated and the distribution of words of different categories is more distinct, the information quality of the text is higher. Therefore, in this embodiment, a method of clustering multiple words contained in the text is adopted, and the semantic score of the text is obtained based on the clustering result.

[0073] In this embodiment, the above method can be used to obtain rule scores from the domain of the word and the type of the word, and to obtain semantic scores from the perspective of word distribution.

[0074] On this basis, the rule score and semantic score can be combined to obtain the quality measurement score of the text to be processed.

[0075] The audio information quality measurement method provided in this embodiment can evaluate the quality of the text based on the field and type of words, as well as the distribution of words represented by clustering results, and can achieve measurement based on the substantive content of the text to obtain measurement results that can represent the quality of the substantive content information.

[0076] In this embodiment, the text to be processed obtained through the processed audio recognition contains multiple sentences. Since the text after audio recognition is colloquial and is affected by the recognition efficiency, some texts in the obtained text to be processed may be too short or too long.

[0077] For example, if the audio is pure music content or the content is misrecognized, the length of the obtained text may be too short, for example, the sentences included may be less than 10. For another example, if a piece of audio is audio in a transition scene, or meaningless audio expressed randomly, or misrecognized, the length of the obtained text may also be short, for example, the sentences included may be less than 15. In the case of such short text, it is difficult to effectively measure the text, so such text can be filtered.

[0078] In addition, if the acquired audio is generated in a live broadcast scenario, for example, the length of the converted text may be too long, for example, it may contain more than 300 sentences. However, not all of these sentences may play a role in representing the actual content information, and processing too long text will have a large processing burden. Therefore, please refer to Figure 2 In this embodiment, after obtaining the text to be processed based on the audio to be processed and before performing entity extraction on the text to be processed, the following steps may also be included:

[0079] S201, when the number of sentences contained in the text to be processed exceeds a set number, select target sentences whose length is less than a preset length from among multiple sentences.

[0080] S202: Filter out the selected target sentences from the text to be processed.

[0081] S203: If the number of sentences contained in the filtered text to be processed exceeds the set number, the text to be processed is truncated to retain the text to be processed with the set number of sentences.

[0082] In this embodiment, the set number can be, for example, 150 sentences, 200 sentences, etc. Among the multiple sentences contained in the text to be processed, there may be some sentences that are not helpful for expressing the substantive content, such as sentences whose length is less than a preset length, where the preset length can be, for example, 10, 15, etc. For example, some exclamatory sentences in audio, affirmative and negative answers in question and answer sentences, etc., have shorter corresponding sentence lengths.

[0083] Such short sentences can be filtered out from the text to be processed, thereby avoiding the problem of processing burden caused by processing too many sentences in the text.

[0084] If the number of sentences in the text to be processed after filtering is less than or equal to the set number, the subsequent entity extraction step can be performed. If the number of sentences in the text to be processed after the above filtering process still exceeds the set number, the text to be processed can be further processed.

[0085] In this embodiment, the text to be processed can be truncated to retain a set number of sentences of the text to be processed. For example, the sentences at the end of the text to be processed can be truncated because the content meaning expressed by the audio at the front end of a piece of audio is often greater than the content meaning expressed by the audio at the back end. Therefore, truncating and filtering the sentences at the end of the text can reduce the burden of text processing without affecting the substantive content expressed by the text.

[0086] In this way, only texts containing no more than a set number of sentences can be included, and subsequent processing can be performed on the texts, thereby avoiding the problem of excessive processing burden.

[0087] See also Figure 3 In this embodiment, in the above step of obtaining the score of each word based on the field to which the word belongs and the type of the word, a possible implementation method is:

[0088] S1031, for each of the words, detect whether the field to which the word belongs is the first preset field, if the field to which the word belongs is the first preset field, execute the following step S1032, if the field to which the word belongs does not belong to the first preset field, execute the following step S1033.

[0089] S1032: Set the score of the word to the score corresponding to the first preset field.

[0090] S1033, obtaining the type of the word, and according to the type of the word, setting a score corresponding to the type for the word.

[0091] In this embodiment, the first preset field may be some fields that require a high level of expertise, such as the financial field, the legal field, the field of historical figures, the medical field, etc. For such fields, scores corresponding to such fields may be preset, and the score may be set relatively large, such as 0.05, so as to increase the influence of words in such fields that require a high level of expertise in the quality results.

[0092] After the score corresponding to the first preset field is pre-configured, when it is determined that a certain word belongs to the first preset field, the score of the word can be set to the score corresponding to the first preset field.

[0093] In addition, if the word does not belong to the first preset field, a score can be set for the word according to the type of the word. Similarly, corresponding scores can be set in advance for each type, for example, words such as names of people, places, and organization names that may appear frequently in the audio can be set to the same score, for example, 0.01.

[0094] In order to avoid the situation where the total score of such words is increased due to the excessive frequency of their appearance in the audio, which makes it difficult to effectively reflect the true quality situation, the sum of the scores of words in each category can be set to an upper limit for such words, for example, it can be set to 0.3 to prevent it from having too much impact. That is, when the sum of the scores of words in each category exceeds 0.3, the sum of their scores is 0.3.

[0095] In addition, for words that do not belong to the first preset field, the type of work title can be set with a corresponding score, such as 0.1. The type of proper noun can be set with a corresponding score, such as 0.5. Similarly, the type of work title and proper noun can also be set with an upper limit value to avoid excessive influence from words of the same type, for example, the upper limit value can be set to 1.

[0096] When setting the word score based on the word type, for some word types that may be inaccurately recognized due to the limitation of recognition effectiveness, such as work titles, proper nouns, etc., when setting the word score for these types of words, you can also take into account whether the word recognition classification is accurate, so as to set their scores accordingly. Optionally, please refer to Figure 4 The above step of setting a score corresponding to the type of the word can be implemented in the following way:

[0097] S10331, check whether the length of the word is greater than or equal to the set length. If it is greater than or equal to the set length, execute the following step S10332. If the length of the word is less than the set length, execute the following step S10333.

[0098] S10332: Set the score of the word to a first score corresponding to its type.

[0099] S10333: Set the score of the word to a second score corresponding to its type, where the second score is greater than the first score.

[0100] In this embodiment, it is considered that the work title and proper noun may make mistakes in recognition and division due to recognition efficiency. One basis for judging whether the word recognition and division is wrong can be judged by the word length of the word. In other words, when the length of the word is too long, there is a high possibility of recognition and division errors. Therefore, it is possible to detect whether the length of the word exceeds the set length, which can be 8, that is, to detect whether the number of characters contained in the word exceeds 8.

[0101] The first score and second score corresponding to the types such as work titles and proper nouns can be set in advance, wherein the second score is greater than the first score, the first score corresponds to words of the type whose length exceeds the set length, and the second score corresponds to words of the type whose length is less than or equal to the set length.

[0102] In this way, for the word that needs to be set, if the length of the word exceeds the set length, the first score under its type can be set for it; if the length of the word is less than or equal to the set length, the second score under its type can be set for it.

[0103] In this way, there may be work titles, proper nouns, etc. that are incorrectly identified and classified, and their scores will be lower than the scores of work titles and proper nouns that are determined to be accurately identified, thereby avoiding the influence of words with recognition problems due to the impact of recognition effect on the measurement results.

[0104] In this embodiment, through the above method, the field to which the word belongs is first determined, and a corresponding score is set for the word based on the pre-set score of the field to which it belongs. In the case where the word does not belong to the preset field, a pre-set score corresponding to the type is set for the word based on the type of the word. The depth of the professional level of the word represented by the field of the word and the type of the word can be used to obtain the depth of the professional level of the information content of the entire text, thereby reflecting the information quality of the text.

[0105] In this embodiment, in order to avoid the difference between the lengths of different texts, which causes the comprehensive scores of some texts to be too large and affects the measurement of different texts, an upper limit value, such as 50, can be set for the rule score. After the rule score is obtained in the above manner, if the rule score exceeds the upper limit value, the rule score can be limited to the upper limit value.

[0106] In addition, in order to unify the standard, the obtained rule score can also be normalized. The rule score can be normalized to a certain range, which can be [0,10], for example. In a possible implementation, the rule score can be implemented by a linear stretching function as shown below:

[0107]

[0108] In this embodiment, multiple words in the text to be processed can also be clustered to obtain a semantic score based on the clustering result. Figure 5 In the above step S104, when clustering multiple words, a possible implementation method is as follows:

[0109] S1041, obtaining the word vector corresponding to each of the words.

[0110] S1042, calculate the cosine similarity between every two word vectors.

[0111] S1043, clustering the multiple word vectors according to the cosine similarity between every two word vectors.

[0112] In this embodiment, the word vector of each word can be obtained. The word vector is a vector representation of a word mapped to a real number. Open source word embedding (TencentAI) can be introduced to obtain the word vector.

[0113] In this embodiment, a vector library can be pre-built, which contains word vectors of multiple common words. For each word in the text to be processed, the preset vector library can be queried, and if the word vector corresponding to the word is found, the word vector can be extracted and obtained.

[0114] Considering that each entity in the text to be processed is extracted based on the entity extraction tool, due to the recognition efficiency of the entity extraction tool, there may be a situation where the extracted entity does not match the actual usage habits. Therefore, for some words, if there are problems with their recognition and division, the corresponding word vector may not be found in the vector library.

[0115] Based on this consideration, in this embodiment, if the word vector corresponding to the word is not found, the word is re-imported into the entity extraction tool for entity extraction, and a query is performed based on the words corresponding to the re-extracted entity until the corresponding word vector is obtained.

[0116] In this embodiment, the entity extraction method is performed by re-importing the word into the entity extraction tool. Although the word may have been extracted incorrectly before, the information subsequently re-imported into the entity extraction tool is different from that before, so the extraction result may also be different. By re-extracting, the error in extraction may be corrected, so that the word vector corresponding to each word is successfully obtained.

[0117] After finding the word vectors corresponding to each word, the cosine similarity between any two words can be calculated based on the word vectors. The higher the cosine similarity, the higher the semantic similarity between the words. For example, it can be understood that the words belong to the same aspect in the same field. The clustering threshold can be set, such as 0.7 or 0.5, where the word vectors with cosine similarity greater than the clustering threshold can be aggregated into the same cluster, and the word vectors exceeding the clustering threshold will be divided into different clusters.

[0118] In this embodiment, a voting mechanism may be used, that is, if the cosine similarity between a word vector and more than half of the word vectors in a cluster is greater than a clustering threshold, the word vector may be classified into the cluster. In this way, the clustering result may be prevented from being too loose.

[0119] After clustering, multiple words will be divided into multiple clusters, among which there may be clusters containing fewer words, for example, less than or equal to 3. For such clusters, in order to avoid the clustering being too loose, such clusters can be filtered out.

[0120] For the clustering results obtained through the above steps, the semantic score of the text to be processed can be calculated based on the clustering results. Figure 6 In this embodiment, the semantic score can be calculated in the following way:

[0121] S1044: Based on the multiple clusters obtained after clustering, calculate the intra-cluster distance of each cluster and the inter-cluster distance between the multiple clusters.

[0122] S1045: Calculate the semantic score of the text to be processed according to the obtained intra-cluster distance and inter-cluster distance.

[0123] In this embodiment, the Dunn Index may be used as a clustering result indicator, so as to measure the information quality of the text from the distribution of multiple words represented by the clustering result.

[0124] Generally speaking, if the content of the audio is concentrated in a certain field, the audio may provide a concentrated and detailed explanation of the knowledge in that field. In this case, the similarity of the words in the text corresponding to the audio should be high, and when clustering, the intra-cluster distance after clustering should be small. Among them, the intra-cluster distance represents the difference between different words in the same cluster, and the intra-cluster distance can be negatively correlated with the cosine similarity between the words in the cluster.

[0125] In addition, if the content involved in the audio includes multiple fields, or includes multiple aspects of the same field, if there are obvious differences between the fields in the audio, or there are large differences between different aspects of the same field, the information quality of the audio can be considered to be high. Words in different fields, or words in different aspects of the same field, should be divided into different clusters in the clustering results. Therefore, the inter-cluster distance in the clustering results can be used to characterize it, that is, the greater the difference between different fields and the greater the difference between different aspects of the same field, the greater the inter-cluster distance should be.

[0126] The inter-cluster distance may be calculated, for example, by calculating the sum of cosine similarities of words in two different clusters, and the difference between the sums of cosine similarities corresponding to the two clusters may be used as the inter-cluster distance.

[0127] Therefore, in this embodiment, the semantic score of the text to be processed can be obtained based on the intra-cluster distance and the inter-cluster distance in the clustering result.

[0128] After obtaining the semantic score of the text to be processed, the semantic score can also be normalized to a predetermined range [0, 10] to facilitate unified standardization.

[0129] Based on the above, the final quality measurement score of the text to be processed can be obtained based on the rule score and semantic score of the text to be processed. Figure 7 In a possible implementation, this step may include the following sub-steps:

[0130] S1051, detecting whether the domain to which the words contained in the to-be-processed text belong is the second preset domain, if the domain to which the words belong is the second preset domain, executing the following steps S1052, if the domain to which the words belong does not belong to the second preset domain, executing the following steps S1053-S1056.

[0131] S1052: Use the smaller value between the rule score and the semantic score as the quality measurement score, or use the minimum value between the rule score, the semantic score and the set score as the quality measurement score.

[0132] S1053: Calculate the difference between the rule score and the semantic score.

[0133] S1054, detecting whether the difference value is within a preset range, if it is within the preset range, executing the following step S1055, if the difference value is not within the preset range, executing the following step S1056.

[0134] S1055: Taking the average of the rule score and the semantic score as the quality measurement score.

[0135] S1056: Use the rule score as the quality measurement score.

[0136] In this embodiment, the second preset field may be, for example, sports, news, professional, literature, etc. Since the professionalism of fields such as professional, sports, and news may not be very high, the quality of the content information contained in the audio may not be very high accordingly, and therefore, the smaller value of the rule score and the semantic score may be used as the final quality measurement score.

[0137] For example, for fields such as literature, the audio is long and the frequency of repeated words may be high. Therefore, the rule score and semantic score for audio in this field may be high, but in fact, the professional level of audio in this field is not very high. Therefore, a set score can be set for this field, and the set score can be set to be smaller.

[0138] The rule score, semantic score and set score can be compared, and the minimum value among the rule score, semantic score and set score is used as the quality measurement score.

[0139] In addition, if the text to be processed does not belong to the second preset field mentioned above, and the difference between the rule score and the semantic score is within the preset range, such as [-2.5, 2.5], it is considered that the two give similar measurement results. In this case, the average of the two can be taken as the final quality measurement score. If the difference between the two exceeds the preset range, the more conservative rule score can be used as the final quality measurement score.

[0140] See also Figure 8 , is a schematic diagram of an exemplary component of an electronic device provided in an embodiment of the present application, and the electronic device may include a storage medium 110, a processor 120, an audio information quality measurement device 130 and a communication interface 140. In this embodiment, the storage medium 110 and the processor 120 are both located in the electronic device and the two are separately arranged. However, it should be understood that the storage medium 110 may also be independent of the electronic device and may be accessed by the processor 120 through a bus interface. Alternatively, the storage medium 110 may also be integrated into the processor 120, for example, it may be a cache and / or a general register.

[0141] The audio information quality measurement device 130 can be understood as the above-mentioned electronic device, or the processor 120 of the electronic device, or can be understood as a software function module independent of the above-mentioned electronic device or processor 120 that implements the above-mentioned audio quality measurement method under the control of the electronic device.

[0142] like Fig. 9As shown, the above-mentioned audio information quality measurement device 130 may include a recognition module 131, an extraction module 132, a calculation module 133, a clustering module 134 and an acquisition module 135. The functions of each functional module of the audio information quality measurement device 130 are described in detail below.

[0143] The recognition module 131 is used to recognize the audio to be processed and obtain the corresponding text to be processed.

[0144] It can be understood that the identification module 131 can be used to execute the above step S101. The detailed implementation of the identification module 131 can refer to the above content related to step S101.

[0145] The extraction module 132 is used to perform entity extraction on the text to be processed to obtain multiple words belonging to the entity.

[0146] It can be understood that the extraction module 132 can be used to execute the above step S102. The detailed implementation of the extraction module 132 can refer to the above content related to step S102.

[0147] The calculation module 133 is used to obtain the score of each word according to a preset rule based on the field to which each word belongs and the type of each word, and to calculate the rule score of the text to be processed according to the scores of multiple words.

[0148] It can be understood that the calculation module 133 can be used to execute the above step S103. The detailed implementation of the calculation module 133 can refer to the above content related to step S103.

[0149] The clustering module 134 is used to perform clustering processing on the multiple words and obtain a semantic score of the text to be processed based on the clustering result.

[0150] It can be understood that the clustering module 134 can be used to execute the above step S104. The detailed implementation of the clustering module 134 can refer to the above content related to step S104.

[0151] The obtaining module 135 is used to obtain the quality measurement score of the text to be processed according to the rule score and the semantic score.

[0152] It can be understood that the obtaining module 135 can be used to execute the above step S105. The detailed implementation of the obtaining module 135 can refer to the above content related to step S105.

[0153] In a possible implementation, the text to be processed includes a plurality of sentences, and the measuring device may further include a filtering module, wherein the filtering module is used to:

[0154] When the number of sentences contained in the to-be-processed text exceeds a set number, screening out target sentences whose length is less than a preset length from among the multiple sentences;

[0155] Filtering the screened target sentences from the text to be processed;

[0156] If the number of sentences contained in the filtered text to be processed exceeds the set number, the text to be processed is truncated to retain the text to be processed with the set number of sentences.

[0157] In a possible implementation, the calculation module 133 may be used to:

[0158] For each of the words, if the field to which the word belongs is a first preset field, the score of the word is set to a score corresponding to the first preset field;

[0159] If the field to which the word belongs does not belong to the first preset field, obtaining the type of the word;

[0160] According to the type of the word, a score corresponding to the type is set for the word.

[0161] In a possible implementation, the calculation module 133 may be used to:

[0162] Detecting whether the length of the word is greater than or equal to a set length, and if so, setting the score of the word to a first score corresponding to its type;

[0163] If the length of the word is less than the set length, the score of the word is set to a second score corresponding to its type, and the second score is greater than the first score.

[0164] In a possible implementation, the clustering module 134 may be used to:

[0165] Obtain word vectors corresponding to each of the words;

[0166] Calculate the cosine similarity between every two word vectors;

[0167] Multiple word vectors are clustered according to the cosine similarity between every two word vectors.

[0168] In a possible implementation, the clustering module 134 may be used to:

[0169] For each of the words, a preset vector library is queried. If the word vector corresponding to the word is not found, the word is re-imported into the entity extraction tool for entity extraction, and a query is performed based on the words corresponding to the re-extracted entity until the corresponding word vector is obtained.

[0170] In a possible implementation, the clustering module 134 may be used to:

[0171] Based on the multiple clusters obtained after clustering, the intra-cluster distance of each cluster and the inter-cluster distance between multiple clusters are calculated;

[0172] The semantic score of the text to be processed is calculated based on the obtained intra-cluster distance and inter-cluster distance.

[0173] In a possible implementation, the obtaining module 135 may be used to:

[0174] When the field to which the words contained in the text to be processed belong is the second preset field, the smaller value between the rule score and the semantic score is used as the quality measurement score, or the minimum value among the rule score, the semantic score and the set score is used as the quality measurement score;

[0175] When the field to which the words contained in the text to be processed belong does not belong to the second preset field, calculating the difference between the rule score and the semantic score;

[0176] If the difference falls within a preset range, the average of the rule score and the semantic score is used as the quality measurement score; if the difference does not fall within the preset range, the rule score is used as the quality measurement score.

[0177] For descriptions of the processing flow of each module in the device and the interaction flow between each module, reference may be made to the relevant descriptions in the above method embodiment, which will not be described in detail here.

[0178] Furthermore, an embodiment of the present application also provides a computer-readable storage medium, which stores machine-executable instructions. When the machine-executable instructions are executed, the audio information quality measurement method provided in the above embodiment is implemented.

[0179] Specifically, the computer-readable storage medium can be a general storage medium, such as a mobile disk, a hard disk, etc. When the computer program on the computer-readable storage medium is executed, the above-mentioned audio information quality measurement method can be executed. Regarding the process involved when the computer-readable storage medium and its executable instructions are executed, reference can be made to the relevant description in the above-mentioned method embodiment, which will not be described in detail here.

[0180] In summary, the audio information quality measurement method, device and electronic device provided in the embodiments of the present application, after obtaining the text to be processed corresponding to the audio to be processed, perform entity extraction on the text to be processed to obtain multiple words belonging to the entity, and obtain the score of each word according to the preset rules based on the field to which each word belongs and the type of each word, and obtain the rule score of the text to be processed. In addition, clustering is performed on the multiple words, and the semantic score of the text to be processed is obtained based on the clustering result, and the quality measurement score of the text to be processed is obtained according to the rule score and the semantic score. In this scheme, the quality evaluation of the text is performed based on the field and type of the words, as well as the distribution of the words represented by the clustering results, so as to achieve the measurement based on the substantive content of the text and obtain the measurement result that can represent the quality of the substantive content information.

[0181] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A method for measuring audio information quality, characterized in that: The method comprises: Recognize the audio to be processed and obtain the corresponding text to be processed; Performing entity extraction on the text to be processed to obtain multiple words belonging to the entity; Based on the fields to which each of the words belongs and the types of each of the words, the scores of each of the words are obtained according to preset rules, and the rule scores of the text to be processed are calculated based on the scores of the multiple words; Performing clustering processing on the multiple words, and obtaining a semantic score of the text to be processed based on the clustering result; Obtaining a quality measurement score of the text to be processed according to the rule score and the semantic score; The step of obtaining the quality measurement score of the text to be processed according to the rule score and the semantic score includes: When the field to which the words contained in the text to be processed belong is the second preset field, the smaller value between the rule score and the semantic score is used as the quality measurement score, or the minimum value between the set score, the rule score and the semantic score is used as the quality measurement score; When the field to which the words contained in the text to be processed belong does not belong to the second preset field, calculating the difference between the rule score and the semantic score; If the difference falls within a preset range, the average of the rule score and the semantic score is used as the quality measurement score; if the difference does not fall within the preset range, the rule score is used as the quality measurement score.

2. The audio information quality measurement method according to claim 1, characterized in that: The text to be processed includes a plurality of sentences. Before the step of extracting entities from the text to be processed, the method further includes: When the number of sentences contained in the to-be-processed text exceeds a set number, screening out target sentences whose length is less than a preset length from among the multiple sentences; Filtering the selected target sentences from the text to be processed; If the number of sentences contained in the filtered text to be processed exceeds the set number, the text to be processed is truncated to retain the text to be processed with the set number of sentences.

3. The audio information quality measurement method according to claim 1, characterized in that: The step of obtaining the score of each word according to a preset rule based on the field to which each word belongs and the type of each word comprises: For each of the words, if the field to which the word belongs is a first preset field, setting the score of the word to a score corresponding to the first preset field; If the field to which the word belongs does not belong to the first preset field, obtaining the type of the word; According to the type of the word, a score corresponding to the type is set for the word.

4. The audio information quality measurement method according to claim 3, characterized in that: The step of setting a score corresponding to the type for the word according to the type of the word comprises: Detecting whether the length of the word is greater than or equal to a set length, and if so, setting the score of the word to a first score corresponding to its type; If the length of the word is less than the set length, the score of the word is set to a second score corresponding to its type, and the second score is greater than the first score.

5. The audio information quality measurement method according to claim 1, characterized in that: The step of clustering the multiple words comprises: Obtain word vectors corresponding to each of the words; Calculate the cosine similarity between every two word vectors; Multiple word vectors are clustered according to the cosine similarity between every two word vectors.

6. The audio information quality measurement method according to claim 5, characterized in that: The step of obtaining the word vector corresponding to each of the words comprises: For each of the words, a preset vector library is queried. If the word vector corresponding to the word is not found, the word is re-imported into the entity extraction tool for entity extraction, and a query is performed based on the words corresponding to the re-extracted entity until the corresponding word vector is obtained.

7. The audio information quality measurement method according to claim 1, characterized in that: The step of obtaining the semantic score of the text to be processed based on the clustering result includes: Based on the multiple clusters obtained after clustering, the intra-cluster distance of each cluster and the inter-cluster distance between multiple clusters are calculated; The semantic score of the text to be processed is calculated based on the obtained intra-cluster distance and inter-cluster distance.

8. An audio information quality measurement device, characterized in that: Used to implement the audio information quality measurement method according to any one of claims 1 to 7, the device comprises: A recognition module is used to recognize the audio to be processed and obtain the corresponding text to be processed; An extraction module, used to perform entity extraction on the text to be processed to obtain multiple words belonging to the entity; A calculation module, used to obtain the score of each word according to a preset rule based on the field to which each word belongs and the type of each word, and to calculate the rule score of the text to be processed according to the scores of multiple words; A clustering module, used for clustering the multiple words and obtaining a semantic score of the text to be processed based on the clustering result; The obtaining module is used to obtain the quality measurement score of the text to be processed according to the rule score and the semantic score.

9. An electronic device, characterized in that: The electronic device comprises one or more storage media and one or more processors communicating with the storage media, wherein the one or more storage media store machine executable instructions executable by the processor, and when the electronic device is running, the processor executes the machine executable instructions to perform the method steps described in any one of claims 1 to 7.

Images