Abnormal voice data detection method and device and related equipment

Abstract

The application discloses an abnormal voice data detection method and device, computer equipment and a storage medium, and applies to the technical field of voice detection. The method comprises the following steps: obtaining historical voice data, obtaining a first corpus set according to the historical voice data, obtaining a second corpus set, and obtaining a representative word set according to the second corpus set and the first corpus set. When the current voice segment of the i-th service scene is detected, whether at least one abnormal word is contained in the current voice segment is determined according to the representative word set corresponding to the i-th service scene before the current voice segment is sent. When at least one abnormal word is contained in the current voice segment, a preset abnormal processing rule is obtained, and the current voice segment is processed based on the preset abnormal processing rule, so that the detection efficiency of abnormal voice data is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of voice detection, in particular to an abnormal voice data detection method, an abnormal voice data detection device, a computer device and a storage medium. BACKGROUND

[0002] Artificial customer service is the bridge and link between enterprises and customers. In different business scenarios, artificial customers have different requirements and standards for language, such as in the scenario of handling telephone business, artificial customer service needs to use polite language to show respect for customers, but in practice, when facing the difficulties of customers or when encountering difficult problems, artificial customer service inevitably cannot control emotions, and uses insulting language to customers, and for example, in the scenario of telephone debt collection, the tone of the customer service personnel is too friendly or uses insulting language to the customers.

[0003] At present, on the one hand, each customer service personnel is trained by professional language before being on duty to avoid the above-mentioned situation, but the training generally needs a long time, especially in the practical scenario, since the customer service personnel is uncontrollable, it is more difficult to effectively avoid the above-mentioned situation, on the other hand, the artificial is arranged to randomly check the telephone recording to detect abnormal voice, which cannot comprehensively detect the telephone recording and is time-consuming and laborious, and the detection efficiency is low. SUMMARY

[0004] The present application provides an abnormal voice data detection method, device, computer device and storage medium to improve the detection efficiency of abnormal voice data.

[0005] An abnormal voice data detection method comprises:

[0006] Obtaining historical voice data of N business scenarios, and obtaining N first corpus sets according to the N historical voice data, wherein N is a positive integer greater than 0;

[0007] Obtaining N second corpus sets input, and obtaining N representative word sets according to the N second corpus sets and the N first corpus sets, wherein N is a positive integer greater than 0, and the N second corpus sets correspond one-to-one to the N first corpus sets;

[0008] When a current voice segment of an i-th business scenario is detected, before the current voice segment is sent, whether at least one abnormal word is contained in the current voice segment is determined according to the representative word set corresponding to the i-th business scenario;

[0009] When the current voice segment contains at least one abnormal word, a preset abnormal processing rule is obtained, and the current voice segment is processed based on the preset abnormal processing rule.

[0010] An abnormal semantic data detection apparatus comprises:

[0011] A first corpus set acquisition module is configured to acquire historical voice data of N service scenarios, and obtain N first corpus sets according to the N historical voice data, wherein N is a positive integer greater than 0.

[0012] A representative word set acquisition module is configured to acquire N second corpus sets, and obtain N representative word sets according to the N second corpus sets and the N first corpus sets, wherein N is a positive integer greater than 0, and the N second corpus sets correspond to the N first corpus sets one by one.

[0013] A first abnormal word determination module is configured to, when performing abnormal detection on a current voice segment of an i-th service scenario, determine whether the current voice segment contains at least one abnormal word according to a representative word set corresponding to the i-th service scenario before the current voice segment is sent.

[0014] A current voice segment processing module is configured to, when the current voice segment contains at least one abnormal word, acquire a preset abnormal processing rule, and process the current voice segment based on the preset abnormal processing rule.

[0015] A computer device comprises a memory, a processor, and a computer program stored in the memory and running on the processor, and the processor implements the steps of the above abnormal voice data detection method when executing the computer program.

[0016] A computer readable storage medium stores a computer program, and the computer program implements the steps of the above abnormal voice data detection method when executed by a processor.

[0017] The application provides an abnormal voice data detection method and device, a computer device and a storage medium. The method comprises the following steps: obtaining historical voice data of N service scenarios, wherein N is a positive integer greater than 0; obtaining N first corpus sets according to the N historical voice data; obtaining N second corpus sets, and obtaining N representative word sets according to the N second corpus sets and the N first corpus sets, wherein N is a positive integer greater than 0, and the N second corpus sets and the N first corpus sets are in one-to-one correspondence; when performing abnormal detection on a current voice segment of an i-th service scenario, determining whether the current voice segment contains at least one abnormal word according to a representative word set corresponding to the i-th service scenario before the current voice segment is sent; and when the current voice segment contains at least one abnormal word, obtaining a preset abnormal processing rule, and processing the current voice segment based on the preset abnormal processing rule. In the application, the representative word set corresponding to each service scenario is obtained from the historical voice data of multiple service scenarios, and it is determined whether the current voice segment corresponding to each service scenario contains at least one abnormal word according to the representative word set corresponding to each service scenario. If the current voice segment contains at least one abnormal word, the current voice segment is processed by using the preset abnormal processing rule, thereby improving the detection efficiency of abnormal voice data of different service scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the description of the embodiments of the application. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0019] Figure 1 is an application environment schematic diagram of an abnormal voice data detection method in an embodiment of the application;

[0020] Figure 2 is a flowchart of an abnormal voice data detection method in an embodiment of the application;

[0021] Figure 3 is a structure schematic diagram of an abnormal voice data detection device in an embodiment of the application;

[0022] Figure 4 is a schematic diagram of a computer device in an embodiment of the application;

[0023] Figure 5 is a flowchart of an abnormal voice data detection method in another embodiment of the application;

[0024] Figure 6 is a flowchart of an abnormal voice data detection method in another embodiment of the application;

[0025] Figure 7 This is a flowchart of an abnormal voice data detection method in another embodiment of this application. Detailed Implementation

[0026] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0027] The abnormal voice data detection method provided in this application embodiment can be applied to, for example, Figure 1 In application environments, such as Figure 1 As shown, system framework 100 may include terminal devices, a network, and a server. The network serves as the medium for providing a communication link between the terminal devices and the server. The network may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0028] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc.

[0029] Terminal devices 101, 102, and 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, e-book readers, MP3 players (Moving Picture Eperts Group Audio Layer III), MP4 players (Moving Picture Eperts Group Audio Layer IV), laptops, and desktop computers, etc.

[0030] Server 105 can be a server that provides various services, such as a backend server that supports the pages displayed on terminal devices 101, 102, and 103.

[0031] It should be noted that the abnormal semantic data detection method provided in this application embodiment is executed by the server, and correspondingly, the abnormal voice data detection device is set in the server.

[0032] It should be understood that Figure 1The number of terminal devices, networks and servers in the embodiments of the present application is only illustrative, and any number of terminal devices, networks and servers can be provided according to actual needs. The terminal device in the embodiments of the present application can specifically correspond to an application system in actual production.

[0033] In an embodiment, as shown in Figure 2 , an abnormal voice data detection method is provided. The method is described by taking a server in Figure 1 as an example, and includes the following steps S201 to S204.

[0034] Step S201: Obtain historical voice data of N service scenarios, and obtain N first corpus sets according to the N historical voice data, where N is a positive integer greater than 0.

[0035] Specifically, the historical voice data of the N service scenarios can be obtained by obtaining the call recording of the customer service personnel in the N service scenarios. The service scenarios can include marketing services, after-sales services, debt collection services, etc. According to the historical voice data corresponding to each service scenario, a first corpus set corresponding to each service scenario is obtained. The first corpus set corresponding to each service scenario can be a word obtained from the historical voice data corresponding to each service scenario. The category form of the word can include a word, a phrase, a short sentence, etc. Different words, phrases and short sentences can also represent different emotions. The category form of the word is not limited here and can be adjusted according to the needs of the actual application scenario.

[0036] In addition, when obtaining the historical voice data of the N service scenarios, if the historical voice data contains at least two audio segments with different characteristics, a DIA voice segmentation method is used to segment the historical voice data to obtain a grouped audio set. Based on a preset script word table, a target audio is selected from the grouped audio set, and the target audio is used as the historical voice data. The preset script word table contains at least one script word. The script word can be a script keyword or script content commonly used by customer service personnel, such as polite language of customer service personnel, etc.

[0037] Step S202: Obtain N second corpus sets, and obtain N representative word sets according to the N second corpus sets and the N first corpus sets, where N is a positive integer greater than 0, and the N second corpus sets and the N first corpus sets correspond one-to-one.

[0038] Specifically, there is a corresponding standard word in each business scenario, wherein the standard word can be a frequently-used script keyword or script content of the customer service staff, the script keyword or script content can be extracted from the training content of the customer service staff or from the example content of the customer service staff in different business scenarios, a preset script word table is obtained, the preset script word table is taken as the second corpus set, and different second corpus sets can be obtained according to different business scenarios.

[0039] The second corpus set includes at least one standard word, and the standard word can be a script keyword or script content. The first corpus set includes at least one word segmentation. Based on a similarity algorithm, semantic similarity of the word segmentation in the first corpus set and the standard word in the second corpus set under the same business scenario is calculated to obtain a script semantic similarity value. If the script semantic similarity value is greater than a preset script semantic similarity threshold, the word segmentation is added to the second corpus set, combined with the standard word, and the second corpus set is taken as a representative word set. Each business scenario corresponds to each representative word set one by one.

[0040] Optionally, the script keyword or script content in the second corpus set can also be classified by manual according to a preset classification rule. The preset classification rule can be to classify the script keyword or script content according to an emotion category of the script keyword or script content. The emotion category can be anger, friendliness, indifference, irritability, etc. The emotion category is not specifically limited here, and it can be defined according to actual application scenarios. For example, the category form of the script keyword or script content can be a word or a word group. The preset classification rule can be to classify the script keyword and script content according to the category form to obtain a word set and a word group set, and then classify the word set and the word group set according to the emotion category of the script keyword or script content.

[0041] Step S203: When the current voice segment of the i th business scenario is detected for abnormality, before the current voice segment is sent, whether at least one abnormal word is included in the current voice segment is determined according to the representative word set corresponding to the i th business scenario.

[0042] Specifically, the current voice segment is voice data of the customer service staff in a preset time period, and the content can be a word, a word, a word group, a sentence, or a paragraph. By performing voice recognition on the current voice segment, current voice text data corresponding to the current voice segment is obtained, and by performing word segmentation processing on the current voice text data, a current word segmentation list is obtained, the current word segmentation list including at least one keyword. Based on a similarity algorithm, the semantic similarity of the keyword and the representative word in the representative word set corresponding to the i-th business scenario is calculated to obtain a semantic similarity value. If there is at least one semantic similarity value greater than a preset semantic similarity threshold, it is determined that the current voice segment contains at least one abnormal word, wherein i is a natural number.

[0043] Step S204: When the current voice segment contains at least one abnormal word, a preset abnormal processing rule is obtained, and the current voice segment is processed based on the preset abnormal processing rule.

[0044] Specifically, the preset abnormal processing rule can be:

[0045] When performing real-time abnormal detection on the current voice segment, if it is determined that the current voice segment contains at least one abnormal word, the current voice segment is processed until the abnormal processing stops when the abnormal processing stop character in the current voice segment is detected. The abnormal processing of the current voice segment includes the muting processing of the current voice segment in the order of time, and the abnormal processing stop character can be the first period in the current voice segment. For example, in the collection scenario, it is determined that the current voice segment contains angry words, and the current voice segment is immediately muted until the first period in the current voice segment is detected, and the muting of the current voice segment is stopped.

[0046] In this embodiment, an abnormal voice data detection method is provided. N historical voice data of business scenarios are obtained, and N first corpus sets are obtained according to the N historical voice data, wherein N is a positive integer greater than 0; N second corpus sets are obtained, and N representative word sets are obtained according to the N second corpus sets and the N first corpus sets, wherein N is a positive integer greater than 0, and the N second corpus sets and the N first corpus sets correspond one by one; when the current voice segment of the i-th business scenario is detected, whether the current voice segment contains at least one abnormal word is determined according to the representative word set corresponding to the i-th business scenario before the current voice segment is sent; when the current voice segment contains at least one abnormal word, a preset abnormal processing rule is obtained, and the current voice segment is processed based on the preset abnormal processing rule. In this embodiment, the representative words in each business scenario are extracted from the historical voice data of multiple business scenarios, the abnormal words in the current voice segment are determined by the representative words, and the detection efficiency of abnormal voice data is improved.

[0047] In some optional implementations of the embodiment, the step S204 of determining whether the current voice segment contains at least one abnormal word according to the representative word set corresponding to the i-th service scenario includes:

[0048] The current voice segment is subjected to speech recognition to obtain current voice text data.

[0049] Specifically, assuming that the voice content of the current voice segment is "Hello, do you want to know the newly launched value-added service?", the current voice text data obtained is "Hello, do you want to know the newly launched value-added service?".

[0050] The current voice text data is subjected to word segmentation processing to obtain a current word segmentation list, and the current word segmentation list includes at least one keyword.

[0051] Specifically, the current voice text data can be subjected to word segmentation processing according to a preset word segmentation dictionary, and the current word segmentation list is obtained, wherein the preset word segmentation dictionary is obtained according to historical experience data analysis.

[0052] The semantic similarity between the keyword and at least one representative word in the representative word set corresponding to the i-th service scenario is calculated according to a similarity algorithm to obtain a semantic similarity value.

[0053] It should be noted that the similarity algorithm in the present application can be cosine similarity, minimum edit distance, Euclidean distance, etc.

[0054] Whether the current voice segment contains at least one abnormal word is determined according to the semantic similarity value and a preset semantic similarity threshold.

[0055] Specifically, if there is at least one voice similarity value greater than the preset semantic similarity threshold, it is determined that the current voice segment contains at least one abnormal word, and if there is no at least one voice similarity value greater than the preset semantic similarity threshold, the current voice segment is sent, wherein the preset semantic similarity value is obtained according to analysis of historical data.

[0056] In the embodiment, the semantic similarity between the keyword in the current voice segment and the representative word in the representative word set is calculated by the similarity algorithm, which is beneficial to quickly and accurately determine whether there is an abnormal word in the current voice segment, thereby improving the detection efficiency of abnormal voice data.

[0057] In some optional implementations of the embodiment, please refer to Figure 5 The step S201 of obtaining N first corpus sets according to N historical voice data includes the following steps S2010 to S2012:

[0058] Step S2010: performing speech recognition on each historical voice data to obtain N pieces of text data corresponding to the N pieces of historical voice data.

[0059] Specifically, speech recognition can be performed on each historical voice data by using a speech recognition technology to obtain N pieces of text data corresponding to the N pieces of historical voice data. The speech recognition technology can convert a voice signal into corresponding text. In this application, the speech recognition technology is used to convert historical voice data into text data.

[0060] Step S2011: performing word segmentation on each piece of text data to obtain N pieces of word segmented text data, wherein each piece of word segmented text data includes at least one word segment.

[0061] Specifically, word segmentation can be performed on each piece of text data based on a preset word segmentation dictionary to obtain N pieces of word segmented text data. The preset word segmentation dictionary can be obtained based on analysis of historical experience data.

[0062] Step S2012: extracting word segments from each piece of word segmented text data to obtain N first corpus sets.

[0063] In this embodiment, by performing speech recognition on historical voice data to obtain text data, and then performing word segmentation on the text data and extracting word segments to obtain first corpus sets, the first corpus sets can more quickly detect abnormal voice data of the corresponding business scenarios, thereby further improving the detection efficiency of abnormal voice data.

[0064] In some optional implementations of this embodiment, after step S2011, word segmentation is performed on each piece of text data to obtain N pieces of word segmented text data, wherein each piece of word segmented text data includes at least one word segment, and then:

[0065] Based on a preset stop word dictionary, it is determined whether the word segment in the word segmented text data is a meaningless word segment.

[0066] Specifically, the preset stop word dictionary can be obtained based on analysis of historical experience data. A similarity algorithm is used to calculate the semantic similarity between the word segment in the word segmented text data and the stop word in the preset stop word dictionary to obtain a semantic similarity value between the word segment and the stop word. If the semantic similarity value between the word segment and the stop word is greater than a preset semantic similarity threshold, it is determined that the word segment is a meaningless word segment.

[0067] If it is determined that the word segment in the word segmented text data is a meaningless word segment, the word segment is removed from the word segmented text data, otherwise, the word segment is retained.

[0068] In this embodiment, by removing the meaningless word segment from the word segmented text data, the effectiveness of the word segment in the word segmented text data is ensured, and the detection efficiency of abnormal voice data is further improved.

[0069] In some optional implementations of the embodiment, referring to Figure 6 , the first corpus set includes a first representative word set and a second representative word set, and the word segmentation includes the first representative word and the second representative word.

[0070] The N first corpus sets are extracted from each word segmented text data, including the following steps S20120 to S20124:

[0071] Step S20120: Extract the first representative word from each word segmented text data, and count the frequency of the first representative word in each word segmented text data to obtain the frequency of the first representative word in each word segmented text data.

[0072] Specifically, the first representative word can be a word, such as "this house", "company", "suspected", "fraud".

[0073] Step S20121: Sort the first representative word frequency list according to the frequency of the first representative word in each word segmented text data.

[0074] Specifically, the first representative word frequency list can be sorted from high to low according to the frequency of the first representative word in each word segmented text data to obtain the N first representative word frequency list.

[0075] Step S20122: Extract M first representative words from each first representative word frequency list to obtain N first representative word sets, wherein M is a positive integer greater than 0.

[0076] Specifically, the N first representative word sets can be obtained by extracting the first M representative words with high frequency from each first representative word frequency list based on a preset extraction rule.

[0077] Step S20123: Extract the second representative word from the corresponding word segmented text data according to the first representative word in each first representative word set to obtain N second representative word sets.

[0078] Specifically, the second type of representative word can be a phrase, such as "this company", "the company is suspected of", "suspected of fraud", the similarity value of the first type of representative and the second type of representative word in the corresponding segmented text data can be calculated by a similarity algorithm, if the similarity value of the first type of representative and the second type of representative word in the corresponding segmented text data is greater than a preset similarity value, the second type of representative word is extracted from the corresponding segmented text data, and N second type of representative word sets are obtained. Assuming that the first type of representative word is "fraud", the second type of representative word in the corresponding segmented text data is "suspected of fraud", the similarity value of "suspected of fraud" and "fraud" is calculated by a similarity algorithm, if the similarity value of "suspected of fraud" and "fraud" is greater than a preset similarity value, "suspected of fraud" is extracted from the corresponding segmented text data.

[0079] Step S20124: merging the first type of representative word set corresponding to each segmented text data and the second type of representative word set corresponding to the first type of representative word, to obtain N representative word sets.

[0080] In this embodiment, the second type of representative word is extracted from the first type of representative word set in the first representative word set to obtain the second type of representative word set, the first type of representative word set and the second type of representative word set are merged, the number of representative words is expanded, which is conducive to more quickly and accurately detecting abnormal voice data corresponding to the business field, and further improving the efficiency of abnormal voice data detection.

[0081] In some optional implementations of this embodiment, please refer to Figure 7 Step S20123: extracting the second type of representative word from the corresponding segmented text data according to the first type of representative word in each first type of representative word set, to obtain N second type of representative word sets, including the following steps S201230 to S201231:

[0082] Step S201230: calculating the semantic similarity of the first type of representative word in each first type of representative word set and the second type of representative word in the corresponding segmented text data based on a similarity algorithm, to obtain a representative word semantic similarity value.

[0083] Step S201231: if the representative word semantic similarity is greater than a preset representative word semantic similarity threshold, the second type of representative word is extracted from the segmented text data, and N second type of representative word sets are obtained.

[0084] Specifically, the representative word voice similarity threshold is obtained according to historical experience data analysis.

[0085] In the embodiment, the second representative word set is obtained by extracting the second representative words from the first representative word set, which effectively expands the number of representative words, and is conducive to more quickly and accurately detecting abnormal voice data in the corresponding business field, and further improves the efficiency of detecting abnormal voice data.

[0086] In some optional implementation of the embodiment, the second representative words include the preposition word group and the postposition word group.

[0087] The semantic similarity between each first representative word in the first representative word set and the second representative word in the corresponding segmented text data is calculated based on the similarity algorithm, and the representative word semantic similarity value is obtained.

[0088] The similarity between each first representative word in the first representative word set and the preposition word group corresponding to the second representative word in the corresponding segmented text data group is calculated, and the representative word semantic similarity is obtained.

[0089] The similarity between each first representative word in the first representative word set and the postposition word group corresponding to the second representative word in the corresponding segmented text data group is calculated, and the representative word semantic similarity is obtained.

[0090] Optionally, the similarity between each first representative word in the first representative word set and the postposition word group corresponding to the second representative word in the corresponding segmented text data group is calculated, and the representative word semantic similarity is obtained. If the representative word semantic similarity value exceeds the preset similarity threshold, the preposition word group of the second representative word is added to the corresponding first representative word set, and the number of first representative words in the first representative word set is expanded, which is conducive to improving the detection efficiency of abnormal voice data.

[0091] For example, assuming that the representative word in the first representative word set is "haha", and the second representative word is "randomly haha", wherein "randomly" is the preposition word group and "haha" is the postposition word group. The semantic similarity between "haha" and "randomly" is calculated based on the similarity algorithm, and the representative word similarity value is obtained. The semantic similarity between "haha" and "haha" is calculated based on the similarity algorithm, and the representative word semantic similarity is obtained. When the semantic similarity between "haha" and "haha" is calculated based on the similarity algorithm, and the representative word semantic similarity is obtained, and the representative word semantic similarity value is greater than the expected similarity threshold, "randomly" is added to the corresponding first representative word set.

[0092] In the embodiment, by calculating the semantic similarity between the first representative word in the first representative word set and the preposition word group of the second representative word and the postposition word group of the second representative word, the number of representative words is expanded, which is conducive to improving the detection efficiency of abnormal voice data.

[0093] In some optional implementations of this embodiment, in step S202, each second corpus set corresponds to a business scenario, the second corpus set includes at least one standard word of the business scenario, each standard word is marked with a preset tag, and the first corpus set includes at least one word segmentation of the business scenario.

[0094] Based on N second corpus sets and N first corpus sets, N representative word sets are obtained, where N is a positive integer greater than 0. The N second corpus sets correspond one-to-one with the N first corpus sets, including:

[0095] Calculate the semantic similarity between the standard words in N second corpora and the corresponding word segments in the first corpus to obtain the word similarity value.

[0096] If the word similarity value is greater than the preset word similarity value, the word segmentation is marked with the preset label of the standard word tag, and the marked word segmentation is added to the second corpus set. The second corpus set is then used as the representative word set to obtain N representative word sets.

[0097] Specifically, preset tags can be emotion tags, such as anger tags or friendliness tags. Standard words and word segments can be classified by emotion tags, thereby enabling faster and more accurate detection of abnormal speech data based on the emotion tags carried by the standard words and word segments.

[0098] In this embodiment, the preset tags carried by the standard words in the second corpus are used to mark the word segments in the first corpus. This is beneficial for classifying the standard words and word segments in business scenarios, thereby enabling faster and more accurate detection of abnormal speech data based on the categories of the standard words and word segments, and improving the efficiency of abnormal speech data detection.

[0099] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0100] In one embodiment, an abnormal data detection device is provided, which corresponds one-to-one with the abnormal data detection methods described in the above embodiments. For example... Figure 3 As shown, the abnormal data detection device includes a first corpus acquisition module 30, a representative word acquisition module 31, a first abnormal word determination module 32, and a current speech segment processing module 33. Detailed descriptions of each functional module are as follows:

[0101] The first corpus set acquisition module 30 is configured to acquire historical voice data of N service scenarios, and obtain N first corpus sets according to the N historical voice data, where N is a positive integer greater than 0.

[0102] The representative word set acquisition module 31 is configured to acquire N second corpus sets input by a user, and obtain N representative word sets according to the N second corpus sets and the N first corpus sets, where N is a positive integer greater than 0, and the N second corpus sets correspond to the N first corpus sets in a one-to-one manner.

[0103] The first abnormal word determination module 32 is configured to, when performing abnormality detection on a current voice segment of an i-th service scenario, determine whether the current voice segment contains at least one abnormal word according to a representative word set corresponding to the i-th service scenario before the current voice segment is sent.

[0104] The current voice segment processing module 33 is configured to, when the current voice segment contains at least one abnormal word, acquire a preset abnormal processing rule, and process the current voice segment based on the preset abnormal processing rule.

[0105] Optionally, the first abnormal word determination module 32 includes:

[0106] The current voice text data acquisition submodule is configured to perform voice recognition on the current voice segment to obtain current voice text data.

[0107] The current word segmentation list acquisition submodule is configured to perform word segmentation processing on the current voice text data to obtain a current word segmentation list, and the current word segmentation list includes at least one keyword.

[0108] The semantic similarity value calculation submodule is configured to calculate a semantic similarity between the keyword and at least one representative word in the representative word set corresponding to the i-th service scenario according to a similarity algorithm to obtain a semantic similarity value.

[0109] The second abnormal word determination submodule is configured to determine whether the current voice segment contains at least one abnormal word according to the semantic similarity value and a preset semantic similarity threshold.

[0110] Optionally, the first corpus set acquisition module 30 includes:

[0111] The text data acquisition submodule is configured to perform voice recognition on each historical voice data to obtain N text data corresponding to the N historical voice data.

[0112] The word segmentation text data acquisition submodule is configured to perform word segmentation processing on each text data to obtain N word segmentation text data, and the word segmentation text data includes at least one word segmentation.

[0113] The word segmentation extraction submodule is configured to extract the word segmentation from each word segmentation text data to obtain N first corpus sets.

[0114] Optionally, the abnormal voice data detection apparatus further comprises:

[0115] The meaningless word segmentation determination module is configured to determine whether the word segmentation in the word segmentation text data is meaningless word segmentation based on the preset stop word dictionary.

[0116] The word segmentation elimination module is configured to eliminate the word segmentation from the word segmentation text data if the word segmentation in the word segmentation text data is determined to be meaningless word segmentation, or to retain the word segmentation.

[0117] Optionally, the first corpus set comprises a first representative word set and a second representative word set, and the word segmentation comprises a first representative word and a second representative word; the first corpus set acquisition module 30 comprises:

[0118] The first representative word extraction submodule is configured to extract the first representative word from each word segmentation text data and perform word frequency statistics on the first representative word in each word segmentation text data to obtain the occurrence frequency of the first representative word in each word segmentation text data.

[0119] The first representative word frequency list obtaining submodule is configured to sort the occurrence frequency of the first representative word in each word segmentation text data to obtain N first representative word frequency lists.

[0120] The first representative word set obtaining submodule is configured to extract M first representative words from each first representative word frequency list to obtain N first representative word sets, wherein M is a positive integer greater than 0.

[0121] The second representative word set acquisition submodule is configured to extract the second representative word from the corresponding word segmentation text data according to the first representative word in each first representative word set to obtain N second representative word sets.

[0122] The representative word set merging submodule is configured to merge the first representative word set corresponding to each word segmentation text data and the second representative word set corresponding to the first representative word to obtain N representative word sets.

[0123] Optionally, the second representative word set acquisition submodule comprises:

[0124] The representative word semantic similarity value calculation unit is configured to calculate the semantic similarity between the first representative word in each first representative word set and the second representative word in the corresponding word segmentation text data based on a similarity algorithm to obtain a representative word semantic similarity value.

[0125] The second-type representative word extraction unit is configured to extract second-type representative words from the segmented text data if the representative word semantic similarity is greater than a preset representative word semantic similarity threshold, to obtain N second-type representative word sets.

[0126] Optionally, the second-type representative words include a preceding word group and a following word group; and the representative word semantic similarity value calculation unit includes:

[0127] The first calculation subunit is configured to calculate the similarity between the first-type representative word in each first-type representative word set and the corresponding preceding word group of the second-type representative word in the corresponding segmented text data set, to obtain the representative word semantic similarity.

[0128] The second calculation subunit is configured to calculate the similarity between the first-type representative word in each first-type representative word set and the corresponding following word group of the second-type representative word in the corresponding segmented text data set, to obtain the representative word semantic similarity.

[0129] Optionally, each second corpus set corresponds to a business scenario, the second corpus set includes at least one standard word of the business scenario, each standard word is marked with a preset label, and the first corpus set includes at least one segmented word of the business scenario; and the representative word set acquisition module 31 includes:

[0130] The third calculation subunit is configured to calculate the semantic similarity between the standard word in the N second corpus sets and the segmented word in the corresponding first corpus set, to obtain a word semantic similarity value.

[0131] The marking subunit is configured to mark the segmented word with the preset label marked by the standard word if the word semantic similarity value is greater than a preset word semantic similarity value, and add the marked segmented word to the second corpus set, to obtain N representative word sets.

[0132] The terms "first" and "second" in the above modules / units are only used to distinguish different modules / units, and do not limit the priority of the modules / units or other limiting meanings. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or modules does not have to include only those steps or modules clearly listed, but can include other steps or modules not clearly listed or inherent to the process, method, product or device. The division of modules in this application is only a logical division, and other division methods can be used in actual application.

[0133] The specific limitations of the abnormal data detection apparatus can refer to the limitations of the abnormal data detection method in the foregoing, which will not be repeated here. Each module in the above abnormal data detection apparatus can be realized by software, hardware, and a combination thereof, in whole or in part. The above modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to each of the above modules.

[0134] In one embodiment, a computer device, which can be a server, is provided, and an internal structure diagram of the computer device can be as shown in Figure 4 The computer device includes a processor, a memory, a network interface, and a database connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The database of the computer device is configured to store data related to the abnormal data detection method. The network interface of the computer device is configured to communicate with an external terminal through a network connection. The computer program is executed by the processor to implement an abnormal data detection method.

[0135] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and running on the processor. The processor implements the steps of the abnormal data detection method in the above embodiments when executing the computer program, such as Figure 2 the steps 201 to 204 and other extensions and related steps of the method as shown in Figure 3 . Alternatively, the processor implements the functions of each module / unit of the abnormal data detection apparatus in the above embodiments when executing the computer program, such as the functions of the modules 30 to 33 as shown in

[0136] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, 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, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The processor is a control center of the computer device, and connects all parts of the computer device through various interfaces and lines.

[0137] The memory can be used to store computer programs and / or modules, and the processor realizes various functions of the computer device by running or executing the computer programs and / or modules stored in the memory, and calling data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application program required by a function (such as a sound playing function, an image playing function, etc.), etc.; and the data storage area can store data created according to the use of the mobile phone (such as audio data, video data, etc.), etc.

[0138] The memory can be integrated in the processor, or can be separately arranged from the processor.

[0139] In an embodiment, a computer readable storage medium is provided, and the computer readable storage medium stores a computer program. The computer program is executed by the processor to realize the steps of the abnormal data detection method in the above embodiment, for example, the steps 201 to 204 and the extension of other extensions and related steps of the method. Figure 2 Or the computer program is executed by the processor to realize the functions of each module / unit of the abnormal data detection device in the above embodiment, for example, the functions of the modules 30 to 33. To avoid repetition, details are not described here. Figure 3 Or the computer program is executed by the processor to realize the functions of each module / unit of the abnormal data detection device in the above embodiment, for example, the functions of the modules 30 to 33. To avoid repetition, details are not described here.

[0140] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing relevant hardware, and the computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, the processes of the above-mentioned embodiments of each method can be included. Any reference to memory, storage, database or other medium used in each embodiment provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0141] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above-mentioned division of each functional unit and module is exemplified, and in actual application, the above-mentioned functions can be completed by different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the above-described functions.

[0142] The above embodiments are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. An abnormal voice data detection method characterized by comprising: The method comprises the following steps: obtaining N sets of historical voice data of service scenarios, and obtaining N first corpus sets according to the N sets of historical voice data: performing voice recognition on each set of historical voice data to obtain N sets of text data corresponding to the N sets of historical voice data; performing word segmentation processing on each set of text data to obtain N sets of word-segmented text data, wherein each set of word-segmented text data comprises at least one word segment, and each word segment comprises a first representative word and a second representative word; extracting the first representative word from each set of word-segmented text data, and counting the frequency of the first representative word in each set of word-segmented text data to obtain the frequency of the first representative word in each set of word-segmented text data; sorting the frequency of the first representative word in each set of word-segmented text data to obtain a first representative word frequency list; extracting M first representative words from each first representative word frequency list to obtain N sets of first representative words, wherein M is a positive integer greater than 0; calculating the similarity between the first representative word in each set of first representative words and the second representative word included in the front word group of the second representative word to obtain a representative word semantic similarity; and calculating the similarity between the first representative word in each set of first representative words and the second representative word included in the rear word group of the second representative word to obtain the representative word semantic similarity; if the representative word semantic similarity is greater than a preset representative word semantic similarity threshold, extracting the second representative word from the word-segmented text data to obtain N sets of second representative words, and the first corpus set comprises the set of first representative words and the set of second representative words, wherein N is a positive integer greater than 0; obtaining N sets of input second corpus sets, and obtaining N sets of representative words according to the N sets of second corpus sets and the N sets of first corpus sets, wherein N is a positive integer greater than 0, and the N sets of second corpus sets correspond one-to-one to the N sets of first corpus sets; when performing anomaly detection on a current voice segment of an i-th service scenario, determining whether at least one abnormal word is included in the current voice segment according to the representative word set corresponding to the i-th service scenario before the current voice segment is sent; when the current voice segment includes at least one abnormal word, obtaining a preset abnormal processing rule, and processing the current voice segment based on the preset abnormal processing rule. 2.The method of claim 1, wherein, After the word segmentation processing on each set of text data to obtain N sets of word-segmented text data, wherein each set of word-segmented text data comprises at least one word segment, and each word segment comprises a first representative word and a second representative word, the method further comprises the following steps: determining whether the word segment in the word-segmented text data is a meaningless word segment based on a preset stop word dictionary; if it is determined that the word segment in the word-segmented text data is a meaningless word segment, the word segment is removed from the word-segmented text data, otherwise, the word segment is retained. 3.The method of claim 1, wherein, Each of the second corpus sets corresponds to one of the business scenarios, and each of the second corpus sets comprises at least one standard word of the business scenario, each of the standard words being marked with a preset label, and each of the first corpus sets comprises at least one segmented word of the business scenario; The N representative word sets are obtained according to the N second corpus sets and the N first corpus sets, wherein N is a positive integer greater than 0, and the N second corpus sets and the N first corpus sets correspond to each other, including: The semantic similarity between the standard word in the N second corpus set and the segmented word in the first corpus set corresponding to the standard word is calculated to obtain a word-phrase similarity value; If the word-phrase similarity value is greater than a preset word-phrase similarity value, the segmented word is marked with the preset label of the standard word, and the marked segmented word is added to the second corpus set, and the second corpus set is taken as the representative word set to obtain the N representative word sets.

4. An abnormal voice data detection apparatus characterized by comprising: The apparatus comprises: The first corpus set acquisition module is configured to acquire historical voice data of N business scenarios, and obtain N first corpus sets according to the N historical voice data: performing voice recognition on each of the historical voice data to obtain N text data corresponding to the N historical voice data; performing segmentation processing on each of the text data to obtain N segmented text data, wherein each of the segmented text data comprises at least one segmented word, and each of the segmented words comprises a first representative word and a second representative word; extracting the first representative words from each of the segmented text data, and performing word frequency statistics on the first representative words in each of the segmented text data to obtain the frequency of occurrence of the first representative words in each of the segmented text data; sorting the first representative words according to the frequency of occurrence of the first representative words in each of the segmented text data to obtain N first representative word frequency lists; extracting M first representative words from each of the first representative word frequency lists to obtain N first representative word sets, wherein M is a positive integer greater than 0; calculating the similarity between the first representative words in each of the first representative word sets and a group of preceding words included in the second representative words in the segmented text data group corresponding to the first representative words to obtain representative word semantic similarity; and calculating the similarity between the first representative words in each of the first representative word sets and a group of following words included in the second representative words in the segmented text data group corresponding to the first representative words to obtain the representative word semantic similarity; if the representative word semantic similarity is greater than a preset representative word semantic similarity threshold, extracting the second representative words from the segmented text data to obtain N second representative word sets, and each of the first corpus sets comprises the first representative word sets and the second representative word sets, wherein N is a positive integer greater than 0. The representative word set acquisition module is configured to acquire N second corpus sets, and obtain N representative word sets according to the N second corpus sets and the N first corpus sets, where N is a positive integer greater than 0, and the N second corpus sets correspond to the N first corpus sets one by one. The first abnormal word determination module is configured to, when performing abnormal detection on a current speech segment of an i th business scenario, determine whether the current speech segment contains at least one abnormal word according to the representative word set corresponding to the i th business scenario before the current speech segment is sent. The current speech segment processing module is configured to, when the current speech segment contains at least one abnormal word, acquire a preset abnormal processing rule, and process the current speech segment based on the preset abnormal processing rule.

5. A computer device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, The processor executes the computer program to implement the steps of the abnormal speech data detection method in any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the abnormal speech data detection method in any one of claims 1 to 3.

Images