CN116705031A - Violation audio detection method and device, electronic equipment, storage medium - Google Patents

Abstract

The embodiment of the application discloses a method and a device for detecting illegal audio, electronic equipment and a storage medium, wherein the method for detecting illegal audio comprises the following steps: acquiring audio to be detected, extracting voice audio in the audio to be detected, and converting the voice audio into text information; inputting the text information into a violation voice detection module to detect whether the violation voice exists in the audio to be detected; the illegal voice detection module detects at least one of voice words in the text information and duration of the audio to be detected; if the voice frequency to be detected has the offensive voice, inputting the text information into a preset offensive detection model to detect whether the voice frequency to be detected is the offensive voice frequency. The technical scheme of the embodiment of the application can rapidly and accurately detect the illegal audio, and simultaneously saves labor cost and time cost.

Description

Violation audio detection method and device, electronic equipment, storage medium

technical field

The present application relates to the technical field of artificial intelligence, in particular, to a method and device for detecting illegal audio, electronic equipment, and a computer-readable storage medium.

Background technique

With the development of multimedia technology, in the network user generated content (User Generated Content, UGC), the user's audio content can be edited and created by himself, or the user can directly adapt the lyrics of the song content, and then spread it on the network, and other users The audio content of its transmission can be viewed on the Internet. However, there may be violations in the audio uploaded to the Internet, and the series of works created by the same singer often have violations.

Therefore, before the audio content is uploaded, it will be detected for violations. Most of the existing violation detections are based on the text content to directly judge whether it is a violation, that is, the audio content is converted into text by means of speech recognition, and then the audio content is processed through the text. Whether it is a violation or not, it may be realized by audio similarity matching. However, due to the music tune and background sound in the audio content, under the influence of the music tune and background sound, the detection rate of the existing violation detection is relatively low. Low. Secondly, there are many types of violations that may be involved in the content of the song, the scene is complex, and it may involve multiple languages and dialects, making it very difficult to judge.

At the same time, in the existing technology, the review of audio content is mainly completed by the double review method of machine review and manual review. There is an urgent need for a violation detection technology that can reduce labor costs and time costs and improve the efficiency of audio content review. The program helps the operation audit team to prohibit the release and delete of audio content that damages the health and safety of the network, so as to build a new pan-entertainment ecology that delivers correct values in a healthy way.

Contents of the invention

In order to solve the above technical problems, the embodiments of the present application provide a violation audio detection method and device, electronic equipment, and a computer-readable storage medium, aiming at solving the high labor cost and time cost of the existing violation review, At the same time, there are technical problems with low audio content review efficiency and accuracy.

Other features and advantages of the present application will become apparent from the following detailed description, or in part, be learned by practice of the present application.

According to an aspect of the embodiment of the present application, a method for detecting illegal audio is provided, including:

Obtain the audio to be detected, extract the human voice audio in the audio to be detected, and convert the human voice audio into text information;

Text information is input into the illegal tone detection module to detect whether there is an illegal tone in the audio to be detected; wherein, the illegal tone detection module is based on at least one of the tone words in the text information and the duration of the audio to be detected;

If there is an illegal tone in the audio to be detected, the text information is input into a preset violation detection model to detect whether the audio to be detected is an illegal audio.

In another embodiment, text information is input into the illegal tone detection module to detect whether there is an illegal tone in the audio to be detected, including:

Obtain the duration information of the audio to be detected, and extract the number of modal particles and vocabulary in the text information, as well as the number of text words in the text information;

Calculate the first ratio between the number of modal particles and the number of vocabulary, and calculate the second ratio between the number of words in the text and the duration information;

Detecting whether there is an illegal tone in the audio to be detected according to the first ratio and the second ratio.

In another embodiment, detecting whether there is an illegal tone in the audio to be detected according to the first ratio and the second ratio includes:

matching the first ratio with the first detection threshold to obtain a first matching result, and matching the second ratio with the second detection threshold to obtain a second matching result;

If the first matching result indicates that the first ratio is greater than the first detection threshold, and the second matching result indicates that the second ratio is less than the second detection threshold, it is determined that there is an illegal tone in the audio to be detected;

If the first matching result indicates that the first ratio is less than or equal to the first detection threshold, or the second matching result indicates that the second ratio is greater than or equal to the second detection threshold, it is determined that there is no illegal tone in the audio to be detected.

In another embodiment, before the text information is input into the illegal tone detection module to detect whether there is an illegal tone in the audio to be detected, the method also includes:

Extract the voiceprint features in the human voice audio to obtain the target voiceprint features;

Match the target voiceprint feature with the illegal voiceprint feature in the voiceprint database to obtain the feature matching result;

If the feature matching result indicates that the target voiceprint feature does not match the illegal voiceprint feature, perform the step of inputting the text information into the preset illegal tone detection module to detect whether there is an illegal tone in the audio to be detected.

In another embodiment, the voiceprint feature in the human voice audio is extracted to obtain the target voiceprint feature, including:

Perform segmentation processing on human voice audio to obtain multiple human voice audio segments;

Extract the voiceprint features of each human voice segment through the voiceprint model;

Calculate the average voiceprint feature of multiple voiceprint features, and use the average voiceprint feature as the target voiceprint feature.

In another embodiment, the method also includes:

If there is no illegal tone in the audio to be detected, the text information is input into the preset semantic model for semantic recognition processing, and the semantic recognition result is obtained;

Determine whether the audio to be detected is an illegal audio according to the semantic recognition result.

In another embodiment, after determining whether the audio to be detected is an illegal audio according to the semantic recognition result, the method further includes:

If the audio to be detected is a violation audio, determine the violation category according to the semantic recognition result;

Determine the corresponding processing flow according to the type of violation, and process the audio to be detected through the processing flow.

According to an aspect of the embodiment of the present application, a violation audio detection device is provided, including:

The obtaining module is configured to obtain the audio to be detected, extract the human voice audio in the audio to be detected, and convert the human voice audio into text information;

The first detection module is configured to input text information into the illegal tone detection module to detect whether there is an illegal tone in the audio to be detected; wherein, the illegal tone detection module is based on at least one of the tone words in the text information and the duration of the audio to be detected. detection;

The second detection module is configured to input text information into a preset violation detection model to detect whether the audio to be detected is an illegal tone if there is an illegal tone in the audio to be detected.

According to an aspect of the embodiments of the present application, an electronic device is provided, including: one or more processors; a storage device for storing one or more programs, when one or more programs are executed by one or more processors When executed, the electronic device is made to implement the above illegal audio detection method.

According to one aspect of the embodiments of the present application, there is provided a computer-readable storage medium, on which computer-readable instructions are stored, and when the computer-readable instructions are executed by the processor of the computer, the computer is made to execute the above-mentioned illegal audio detection method .

According to an aspect of the embodiments of the present application, a computer program product or computer program is provided, the computer program product or computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the violation audio detection method provided in the above-mentioned various optional embodiments.

In the technical solution provided in the embodiments of the present application, when the audio to be detected needs to be detected, the human voice audio in the audio to be detected is extracted, and the human voice audio is converted into text information, the extracted human voice audio will be very loud Minimize the interference of background music, thereby improving the recognition accuracy of converting human voice audio into text information. The present application also inputs the converted text information into the illegal tone detection module to detect whether there is an illegal tone in the audio to be detected, and the illegal tone detection module detects at least one of the modal particles in the text information and the duration of the audio to be detected. When there is an illegal tone in the audio to be detected, there are certain differences in the modal particles and duration compared with the normal audio. Therefore, the illegal tone detection module can accurately detect whether there is an illegal tone in the audio to be detected. When it is detected that there is an illegal tone in the audio to be detected, the text information is further input into the preset violation detection model to detect whether the audio to be detected is an illegal audio, and the violation detection can be performed more quickly and accurately. It can be seen that adopting the technical solution of this application does not need to set up a proprietary manual review, which can reduce the labor cost and time cost, and help the operation review team to prohibit the release and deletion of audio content that damages the health and safety of the network, and then build a healthy delivery of correct values. new pan-entertainment ecology.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the application.

Description of drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the application and together with the description serve to explain the principles of the application. Apparently, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings based on these drawings without creative efforts. In the attached picture:

FIG. 1 is a schematic diagram of an implementation environment involved in the present application;

Fig. 2 is the flowchart of a kind of illegal audio detection method involved in the present application;

Fig. 3 is the flow chart of the illegal audio detection method in another embodiment involved in the present application;

FIG. 4 is a flowchart of step S310 in an embodiment involved in the present application;

FIG. 5 is a flowchart of step S220 in an embodiment involved in the present application;

FIG. 6 is a flowchart of step S530 in an embodiment involved in the present application;

Fig. 7 is the flow chart of the illegal audio detection method in another embodiment involved in the present application;

FIG. 8 is a flowchart of step S720 in an embodiment involved in the present application;

Fig. 9 is a flow chart of a method for detecting illegal audio in another embodiment involved in the present application;

Fig. 10 is a block diagram of a violation audio detection device involved in the present application;

FIG. 11 shows a schematic structural diagram of a computer system suitable for implementing the electronic device of the embodiment of the present application.

Detailed ways

Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numerals in different drawings refer to the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with aspects of the present application as recited in the appended claims.

The block diagrams shown in the drawings are merely functional entities and do not necessarily correspond to physically separate entities. That is, these functional entities may be implemented in software, or in one or more hardware modules or integrated circuits, or in different networks and/or processor devices and/or microcontroller devices entity.

The flow charts shown in the drawings are only exemplary illustrations, and do not necessarily include all contents and operations/steps, nor must they be performed in the order described. For example, some operations/steps can be decomposed, and some operations/steps can be combined or partly combined, so the actual order of execution may be changed according to the actual situation.

It should also be noted that: the "plurality" mentioned in this application refers to two or more. "And/or" describes the association relationship of associated objects, indicating that there may be three types of relationships. For example, A and/or B may indicate: A exists alone, A and B exist simultaneously, and B exists independently. The character "/" generally indicates that the contextual objects are an "or" relationship.

Please refer to FIG. 1 , which is a schematic diagram of an implementation environment involved in this application. The implementation environment includes a terminal 110 and a server 120, and the terminal 110 and the server 120 communicate through a wired or wireless network.

Terminal 110 runs applications for creating user-generated content and applications for playing audio and video. In some playing applications, user-generated content can be created directly. When a user uploads user-generated content to an application for playback, if the uploaded application is a public platform, that is, when the user-generated content uploaded by the user can be played by other users, the user-generated content uploaded by the user needs to be reviewed to determine Whether the user-generated content contains illegal content, and when it is determined that the user-generated content contains illegal content, the user is not allowed to successfully upload the user-generated content.

The server 120 is provided with a method for detecting violations of content uploaded by users to implement violation detection for audio, images or videos.

Wherein, the terminal 110 can be any electronic device capable of running authoring applications and playing applications, such as smart phones, tablets, laptops, computers, etc., and the server 120 can be an independent physical server, or a server cluster composed of multiple physical servers Or a distributed system can also provide cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network, content distribution network) and Cloud servers for basic cloud computing services such as big data and artificial intelligence platforms are not restricted here.

Fig. 2 is a flow chart showing a method for detecting illegal audio according to an exemplary embodiment. The method can be applied to the implementation environment shown in FIG. 1 , and is specifically executed by the server 120 in the embodiment environment shown in FIG. 1 .

As shown in FIG. 2, in an exemplary embodiment, the method for detecting illegal audio may include steps S210 to S230, which are described in detail as follows:

Step S210, acquiring the audio to be detected, extracting human voice audio from the audio to be detected, and converting the human voice audio into text information.

In the embodiment of this application, when the user uploads audio, the uploaded audio is used as the audio to be detected, and a detection request is generated. When the detection request is received, the audio to be detected corresponding to the detection request is obtained. The audio to be detected includes background audio , melody audio, human voice audio, etc., extract the human voice audio in the audio to be detected by suppressing and separating the background audio, melody audio, etc., and convert the human voice audio into text information. By suppressing and separating background audio and melody audio, this application will greatly reduce the interference of music on voiceprint features and speech features, and greatly improve the subsequent conversion of human voice audio into text information and the extraction of voice information. Unique recognition accuracy.

Specifically, when extracting human voice audio, the audio to be detected is input into the preset human voice separation model for processing. The human voice separation model can use deep neural network models such as U-Net, Wave-U-Net, WaveGlow, Deep Clustering (deep clustering), TasNet (Time-domain audio separation Network, time-domain audio separation network), Conv-TasNet (Convolutional Time-domain audio separation Network, fully convolutional time-domain audio separation network model), etc., or based on attention It is trained by TasNet and Conv-TasNet with force mechanism. The human voice separation model separates the audio to be detected into two audio files: human voice audio and other audio except the human voice audio. In the embodiment of the present application, only human voice audio is reserved for subsequent processing.

Perform speech recognition processing on the extracted human voice audio to obtain text information. Specifically, human voice audio is input into the pre-trained speech recognition model for speech recognition to obtain text information.

In an exemplary embodiment of the present application, please refer to FIG. 3 , before step S220, the text information is input into the illegal tone detection module to detect whether there is an illegal tone in the audio to be detected, the illegal audio detection method also includes steps S310 to S330 , the details are as follows:

Step S310, extracting voiceprint features in the human voice audio to obtain target voiceprint features.

In the embodiment of the present application, the voiceprint feature is extracted from the human voice audio as the target voiceprint feature. Specifically, it can be extracted through a trained voiceprint model, and the voiceprint model can be obtained by training based on a deep learning TDNN (TimeDelay Neural Network, Time Delay Neural Network) network structure.

Step S320, matching the target voiceprint features with the illegal voiceprint features in the voiceprint library to obtain a feature matching result.

In the embodiment of the present application, voiceprint feature extraction is performed on fragments of illegal audio found in operation or notified in advance, the x-vector of the extracted voiceprint is combined into a voiceprint library, and the publisher corresponding to the illegal audio is Mark as a sensitive person, and use the voiceprint feature of the extracted illegal audio as the voiceprint feature of the violation.

Match the target voiceprint feature with the violation voiceprint feature to obtain the feature matching result. The feature matching result is divided into two categories, one is that the target voiceprint feature does not match the violation voiceprint feature, and the other is that the target voiceprint Voiceprint feature matching.

Step S330, if the feature matching result indicates that the target voiceprint feature does not match the illegal voiceprint feature, perform the step of inputting the text information into the preset illegal tone detection module to detect whether there is an illegal tone in the audio to be detected.

In the embodiment of the present application, if the feature matching result indicates that the target voiceprint feature does not match the illegal voiceprint feature, the step of inputting the text information into the preset illegal tone detection module to detect whether there is an illegal tone in the audio to be detected is performed. If the target voiceprint feature matches the illegal voiceprint feature, there is no need to perform subsequent processing steps of inputting the text information into the illegal tone detection module, and the publisher is directly not allowed to publish the audio to be detected on the application program.

In another embodiment of the present application, when the target voiceprint feature matches the illegal voiceprint feature, if the target voiceprint feature matches the illegal voiceprint feature, the text information can be input into the preset semantic model for semantic recognition processing, and the obtained The semantic recognition result, and then further determine whether the audio to be detected is an illegal audio according to the semantic recognition result.

In the embodiment of this application, through the matching of voiceprint features, it is not necessary for all the audio to be detected to be detected by the violation tone detection module and the violation detection model, which greatly saves the computing resources of the server and improves the detection efficiency of violation detection.

In an exemplary embodiment of the present application, please refer to FIG. 4 , in step S310, the voiceprint feature in the human voice audio is extracted to obtain the target voiceprint feature, including steps S410 to S430, which are described in detail as follows:

Step S410, segmenting the human voice audio to obtain multiple human voice audio segments.

In the embodiment of the present application, the human voice audio is segmented to obtain individual voice segments. During segment processing, the corresponding number of segments can be preset, and the human voice audio can be divided into multiple human voice audio segments according to the fixed number of segments. In another embodiment, the duration of each human voice segment may be preset, and the human voice audio may be cut into fixed duration human voice segments. The aforementioned preset number of segments and duration can be set based on needs, which is not limited in this application.

Step S420, extracting the voiceprint features of each human voice segment through the voiceprint model.

In the embodiment of the present application, each human voice segment is input into the voiceprint model for voiceprint recognition, and the voiceprint feature x-vector of each human voice segment is extracted.

Step S430, calculating the average voiceprint feature of multiple voiceprint features, and using the average voiceprint feature as the target voiceprint feature.

In the embodiment of the present application, multiple voiceprint feature x-vectors are averaged to obtain the average voiceprint feature. The averaged average voiceprint feature is used as the target voiceprint feature to match with the illegal voiceprint feature in the voiceprint library.

Step S220, input the text information into the illegal tone detection module to detect whether there is an illegal tone in the audio to be detected; wherein, the illegal tone detection module detects based on at least one of the modal particles in the text information and the duration of the audio to be detected.

In the embodiment of the present application, an illegal tone detection module is preset, and the illegal tone detection module performs detection based on at least one of the modal particle in the text information and the duration of the audio to be detected. Input the converted text information into the illegal tone detection module to detect whether there is an illegal tone in the audio to be detected. The illegal tone in the embodiment of the present application can be understood as expressing the illegal content through modal particles. Therefore, when there is an illegal tone in the audio to be detected, the modal particles and duration are different from normal audio.

In an exemplary embodiment of the present application, please refer to FIG. 5, in step S220, input text information into the illegal tone detection module to detect whether there is an illegal tone in the audio to be detected, including steps S510 to step S530, described in detail as follows:

In step S510, the duration information of the audio to be detected is obtained, and the number of modal particles and the number of vocabulary in the text information, as well as the number of words in the text information are extracted.

In the embodiment of the present application, the duration information of the audio to be detected is obtained, and word segmentation processing is performed on the text information. When performing word segmentation processing, tools such as stuttering word segmentation, SnowNLP, THULAC (THU Lexical Analyzer for Chinese), NLPIR word segmentation system can be used. Word segmentation processing. During the word segmentation process, perform part-of-speech tagging on each word after word segmentation, and then extract modal particles from the part-of-speech tagged vocabulary, and count the number of modal particles and vocabulary, as well as the number of text words in the text information.

Step S520, calculating the first ratio between the number of modal particles and the number of vocabulary, and calculating the second ratio between the number of words in the text and the duration information.

In the embodiment of the present application, the first ratio is obtained by dividing the number of modal particles by the number of vocabulary, and the second ratio is obtained by dividing the number of words in the text by the duration information.

Step S530, detecting whether there is an illegal tone in the audio to be detected according to the first ratio and the second ratio.

In the embodiment of the present application, usually, when there is an illegal tone in the audio to be detected, the vocabulary of the non-modal particles in the human voice audio is relatively small, while the proportion of the modal particles such as "ah, oh, ha, good" is than greatly increased. In addition, the number of recognized words is much smaller than normal audio under the same audio length, so whether there is an illegal tone in the audio to be detected can be detected according to the first ratio and the second ratio.

In an exemplary embodiment of the present application, referring to FIG. 6 , in step S530, it is detected whether there is an illegal tone in the audio to be detected according to the first ratio and the second ratio, including steps S610 to S630, which are described in detail as follows:

Step S610, matching the first ratio with the first detection threshold to obtain a first matching result, and matching the second ratio with the second detection threshold to obtain a second matching result.

In the embodiment of the present application, the first detection threshold and the second detection threshold are preset, the first ratio is matched with the first detection threshold to obtain the first matching result, and the second ratio is matched with the second detection threshold, Get the second matching result.

Step S620, if the first matching result indicates that the first ratio is greater than the first detection threshold, and the second matching result indicates that the second ratio is less than the second detection threshold, it is determined that there is an illegal tone in the audio to be detected.

In the embodiment of the present application, if the first ratio is greater than the first detection threshold and the second ratio is smaller than the second detection threshold, it indicates that the proportion of non-modal particles in the vocal audio is relatively small, while the proportion of modal particles is relatively large. At the same time, the number of recognized texts is also much smaller than the normal audio under the same audio length, so it can be determined that there is an illegal tone in the audio to be detected.

Step S630, if the first matching result indicates that the first ratio is less than or equal to the first detection threshold, or the second matching result indicates that the second ratio is greater than or equal to the second detection threshold, determine that there is no illegal tone in the audio to be detected.

In the embodiment of the present application, if any of the first ratio is less than or equal to the first detection threshold, or the second ratio is greater than or equal to the second detection threshold, it indicates that the vocabulary of non-modal particles in the human voice audio accounts for a large proportion, while the tone If the proportion of words is small, or the number of recognized words reaches normal audio under the same audio length, it can be determined that there is no illegal tone in the audio to be detected.

Step S230, if there is an illegal tone in the audio to be detected, input text information into a preset violation detection model to detect whether the audio to be detected is an illegal audio.

In the embodiment of the present application, the preset violation detection model can be trained based on the Panns-cnn10 model or the Ecapa-tdnn model. During training, there are two types of positive samples and negative samples. Among them, natural sounds are mostly used in positive samples. Specifically, it may include animal sounds, natural sounds & water sounds, non-conversational human voices, household sounds, urban noises, etc.

Negative samples use labeled offending audio.

After training the Panns-cnn10 model or Ecapa-tdnn through positive and negative samples to obtain the baseline model, migration training was performed, and data such as songs, recorded courses, and audio novels were added during the migration training, making the final training violations The detection model can be more accurate, and the overall training time is shorter. In an embodiment of the present application, the Ecapa-tdnn model is preferably trained to obtain a violation detection model.

After it is determined that there is an illegal tone in the audio to be detected, the text information is input into a preset violation detection model to further detect whether the audio to be detected is an illegal audio, so that violation detection can be performed more accurately.

In the embodiment of this application, if the audio to be detected is detected to be an illegal audio according to the violation detection model, the audio to be detected is input to the manual review channel for further review, and the obtained manual review result indicates that the audio to be detected is a violation audio, then The target voiceprint feature extracted from the audio to be detected is stored in the voiceprint library as the voiceprint feature of the violation, and the publisher of the audio to be detected is marked as a sensitive person.

In the embodiment of the present application, when the audio to be detected needs to be detected, the human voice audio in the audio to be detected is extracted, and the human voice audio is converted into text information, and the extracted human voice audio will greatly reduce background music interference, Thereby, the recognition accuracy of converting human voice audio into text information is improved. The present application also inputs the converted text information into the illegal tone detection module to detect whether there is an illegal tone in the audio to be detected. The illegal tone detection module detects at least one of the modal particles in the text information and the duration of the audio to be detected. When there is an illegal tone in the audio to be detected, the modal particles and duration are different from the normal audio. Therefore, the illegal tone detection module can accurately detect whether there is an illegal tone in the audio to be detected. When it is detected that there is an illegal tone in the audio to be detected, the text information is further input into the preset violation detection model to detect whether the audio to be detected is an illegal audio, so that the violation detection can be performed more quickly and accurately. It can be seen that adopting the technical solution of this application does not need to set up a proprietary manual review, which can reduce the labor cost and time cost, and help the operation review team to prohibit the release and deletion of audio content that damages the health and safety of the network, and then build a healthy delivery of correct values. new pan-entertainment ecology.

In an exemplary embodiment of the present application, please refer to FIG. 7, the illegal audio detection method also includes step S710 and step S720, which are described in detail as follows:

Step S710, if there is no illegal tone in the audio to be detected, input text information into a preset semantic model for semantic recognition processing, and obtain a semantic recognition result.

In the embodiment of the present application, after the illegal tone detection module detects that there is no illegal tone in the audio to be detected, the text information is input into the preset semantic model for semantic recognition processing, and the semantic recognition result is obtained.

Step S720, determine whether the audio to be detected is an illegal audio according to the semantic recognition result.

In the embodiment of the present application, the semantics expressed by the audio to be detected can be determined according to the semantic recognition result, and then whether the audio to be detected is an illegal audio can be determined.

In an exemplary embodiment of the present application, please refer to FIG. 8. After determining whether the audio to be detected is an illegal audio according to the semantic recognition result in step S720, the illegal audio detection method further includes steps S810 and S820, which are described in detail as follows:

Step S810, if the audio to be detected is a violation audio, determine the violation type according to the semantic recognition result.

In the embodiment of the present application, when the audio to be detected is determined to be a violation audio, the violation type is further determined according to the semantic recognition result, and there are many types of violations.

Step S820, determine the corresponding processing flow according to the type of violation, and process the audio to be detected through the processing flow.

In the embodiment of this application, each violation category has a corresponding processing flow, such as prohibited, suspected, in the prohibited processing flow, it is forbidden to publish the audio to be detected to the network, and in the suspected processing flow, further processing is required. Determine if it is the offending audio. The audio to be detected is processed according to the determined processing flow.

In an exemplary embodiment of the present application, please refer to FIG. 9. FIG. 9 is a method for detecting illegal audio according to an exemplary embodiment, including steps S910 to S960. The details are as follows:

Step S910, if a detection request is received, acquire audio to be detected according to the detection request.

In this embodiment of the application, when the user uploads audio, the uploaded audio is used as the audio to be detected, and a detection request is generated, and when the detection request is received, the audio to be detected corresponding to the detection request is obtained.

Step S920, extracting human voice audio from the audio to be detected.

In the embodiment of the present application, the audio to be detected is processed according to the preset human voice separation model, and the human voice audio in the audio to be detected is extracted by suppressing and separating background audio and melody audio.

Step S930, extracting the voiceprint features in the human voice audio to obtain the target voiceprint features, and matching the target voiceprint features with the illegal voiceprint features in the voiceprint database to obtain a feature matching result.

In the embodiment of the present application, step S930 is consistent with the descriptions in step S310 and step S320 described above, and will not be repeated here.

Step S940, if the feature matching result indicates that the target voiceprint feature does not match the illegal voiceprint feature, input the text information into the preset semantic model for semantic recognition processing, obtain the semantic recognition result, and convert the human voice audio into text information.

In the embodiment of the present application, if the feature matching result indicates that the target voiceprint feature does not match the illegal voiceprint feature, the text information is directly input into the preset semantic model for semantic recognition processing, and the semantic recognition result is obtained. At the same time, the human voice audio is converted into text message.

Step S950, if the semantic recognition result indicates that the audio to be detected is not an illegal audio, input the text information into the preset illegal tone detection module to detect whether there is an illegal tone in the audio to be detected; wherein, the illegal tone detection module is based on the tone in the text information At least one of the word and the duration of the audio to be detected is detected.

In the embodiment of the present application, when the semantic recognition result indicates that the audio to be detected is not an illegal audio, the converted text information is input into the preset illegal tone detection module to detect whether there is an illegal tone in the audio to be detected, how does the tone detection module detect The description of whether there is an illegal tone in the audio to be detected is consistent with the description of the aforementioned steps S510 to S530, and steps S610 to S630, and will not be repeated here.

Step S960, if there is an illegal tone in the audio to be detected, input text information into a preset violation detection model to detect whether the audio to be detected is an illegal audio.

In the embodiment of the present application, if there is an illegal tone in the audio to be detected, the text information is input into the preset violation detection model to detect whether the audio to be detected is an illegal audio. The relevant description of step S960 is consistent with the relevant description of the aforementioned step S230. This will not be repeated here.

In an exemplary embodiment of the present application, please refer to FIG. 10. FIG. 10 is a device for detecting illegal audio according to an exemplary embodiment, including:

The acquisition module 1010 is configured to acquire the audio to be detected, extract the human voice audio in the audio to be detected, and convert the human voice audio into text information;

The first detection module 1020 is configured to input text information into the illegal tone detection module to detect whether there is an illegal tone in the audio to be detected; wherein, the illegal tone detection module is based on at least one of the tone words in the text information and the duration of the audio to be detected to test;

The second detection module 1030 is configured to input text information into a preset violation detection model to detect whether the audio to be detected is an illegal tone if there is an illegal tone in the audio to be detected.

In an exemplary embodiment of the present application, the first detection module 1020 includes:

The first extraction submodule is configured to obtain the duration information of the audio to be detected, and extract the number of modal particles and the number of vocabulary in the text information, and the number of words in the text information;

The first calculation submodule is configured to calculate the first ratio between the number of modal particles and the number of vocabulary, and calculate the second ratio between the number of words in the text and the duration information;

The detection submodule is configured to detect whether there is an illegal tone in the audio to be detected according to the first ratio and the second ratio.

In an exemplary embodiment of the present application, the detection submodule includes:

A matching unit configured to match the first ratio with the first detection threshold to obtain a first matching result, and match the second ratio with the second detection threshold to obtain a second matching result;

The first determination unit is configured to determine that there is an illegal tone in the audio to be detected if the first matching result indicates that the first ratio is greater than the first detection threshold, and the second matching result indicates that the second ratio is less than the second detection threshold;

The second determination unit is configured to determine that there is no illegal tone in the audio to be detected if the first matching result indicates that the first ratio is less than or equal to the first detection threshold, or the second matching result indicates that the second ratio is greater than or equal to the second detection threshold.

In an exemplary embodiment of the present application, the illegal audio detection device also includes:

The extraction module is configured to extract the voiceprint feature in the human voice audio to obtain the target voiceprint feature;

The matching module is configured to match the target voiceprint feature with the violation voiceprint feature in the voiceprint library to obtain a feature matching result;

The execution module is configured to execute the step of inputting text information into a preset illegal tone detection module to detect whether there is an illegal tone in the audio to be detected if the feature matching result indicates that the target voiceprint feature does not match the illegal voiceprint feature.

In an exemplary embodiment of the present application, the extraction module includes:

The segmentation processing sub-module is configured to segment the human voice audio to obtain multiple human voice audio segments;

The second extraction submodule is configured to extract the voiceprint features of each human voice segment through the voiceprint model;

The second calculation sub-module is configured to calculate the average voiceprint feature of multiple voiceprint features, and use the average voiceprint feature as the target voiceprint feature.

In an exemplary embodiment of the present application, the illegal audio detection device also includes:

The semantic recognition module is configured to input text information into a preset semantic model for semantic recognition processing if there is no illegal tone in the audio to be detected, and obtain a semantic recognition result;

The first determination module is configured to determine whether the audio to be detected is an illegal audio according to the semantic recognition result.

In an exemplary embodiment of the present application, the illegal audio detection device also includes:

The second determination module is configured to determine the violation category according to the semantic recognition result if the audio to be detected is an illegal audio;

The third determining module is configured to determine a corresponding processing flow according to the violation category, and process the audio to be detected through the processing flow.

It should be noted that the illegal audio detection device provided in the above-mentioned embodiments and the illegal audio detection method provided in the above-mentioned embodiments belong to the same idea, and the specific ways of performing operations of each module, sub-module and unit have been carried out in the method embodiment A detailed description is given and will not be repeated here.

The embodiment of the present application also provides an electronic device, including: one or more processors; a storage device for storing one or more programs, when the one or more programs are processed by the one or more When the device is executed, the electronic device is made to implement the violation audio detection method provided in each of the above embodiments;

Violation audio detection methods, including:

Obtain the audio to be detected, extract the human voice audio in the audio to be detected, and convert the human voice audio into text information;

Text information is input into the illegal tone detection module to detect whether there is an illegal tone in the audio to be detected; wherein, the illegal tone detection module is based on at least one of the tone words in the text information and the duration of the audio to be detected;

If there is an illegal tone in the audio to be detected, the text information is input into a preset violation detection model to detect whether the audio to be detected is an illegal audio.

FIG. 11 shows a schematic structural diagram of a computer system suitable for implementing the electronic device of the embodiment of the present application.

It should be noted that the computer system 1100 of the electronic device shown in FIG. 11 is only an example, and should not limit the functions and scope of use of this embodiment of the present application.

As shown in FIG. 11 , a computer system 1100 includes a central processing unit (Central Processing Unit, CPU) 1101, which can be stored in a program in a read-only memory (Read-Only Memory, ROM) 1102 or loaded from a storage part 1108 to a random It accesses programs in the memory (Random Access Memory, RAM) 1103 to execute various appropriate actions and processes, such as executing the methods described in the above-mentioned embodiments. In the RAM 1103, various programs and data necessary for system operation are also stored. The CPU 1101 , ROM 1102 , and RAM 1103 are connected to each other via a bus 1104 . An input/output (Input/Output, I/O) interface 1105 is also connected to the bus 1104 .

The following components are connected to the I/O interface 1105: an input section 1106 including a keyboard, a mouse, etc.; an output section 1107 including a cathode ray tube (Cathode Ray Tube, CRT), a liquid crystal display (Liquid Crystal Display, LCD), etc., and a speaker ; a storage section 1108 including a hard disk or the like; and a communication section 1109 including a network interface card such as a LAN (Local Area Network) card, a modem, or the like. The communication section 1109 performs communication processing via a network such as the Internet. A drive 1110 is also connected to the I/O interface 1105 as needed. A removable medium 1111 such as a magnetic disk, optical disk, magneto-optical disk, semiconductor memory, etc. is mounted on the drive 1110 as necessary so that a computer program read therefrom is installed into the storage section 1108 as necessary.

In particular, according to the embodiments of the present application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, the embodiments of the present application include a computer program product, which includes a computer program carried on a computer-readable medium, where the computer program includes a computer program for executing the method shown in the flowchart. In such an embodiment, the computer program may be downloaded and installed from a network via communication portion 1109, and/or installed from removable media 1111. When the computer program is executed by a central processing unit (CPU) 1101, various functions defined in the system of the present application are performed.

It should be noted that the computer-readable medium shown in the embodiment of the present application may be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium may be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to, electrical connections with one or more wires, portable computer diskettes, hard disks, random access memory (RAM), read-only memory (ROM), erasable Programmable Read-Only Memory (Erasable Programmable Read Only Memory, EPROM), flash memory, optical fiber, portable compact disk read-only memory (Compact Disc Read-Only Memory, CD-ROM), optical storage device, magnetic storage device, or any suitable The combination. In this application, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, however, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, in which a computer-readable computer program is carried. Such propagated data signals may take many forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. . A computer program embodied on a computer readable medium can be transmitted using any appropriate medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

The flowchart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. Wherein, each block in the flowchart or block diagram may represent a module, a program segment, or a part of the code, and the above-mentioned module, program segment, or part of the code includes one or more executable instruction. It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or they may sometimes be executed in the reverse order, depending upon the functionality involved. It should also be noted that each block in the block diagrams or flowchart illustrations, and combinations of blocks in the block diagrams or flowchart illustrations, can be implemented by a dedicated hardware-based system that performs the specified function or operation, or can be implemented by a A combination of dedicated hardware and computer instructions.

The units described in the embodiments of the present application may be implemented by software or by hardware, and the described units may also be set in a processor. Wherein, the names of these units do not constitute a limitation of the unit itself under certain circumstances.

Another aspect of the present application also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the aforementioned method is implemented. The computer-readable storage medium may be included in the electronic device described in the above embodiments, or may exist independently without being assembled into the electronic device.

Another aspect of the present application also provides a computer program product or computer program, the computer program product or computer program comprising computer instructions stored in a computer-readable storage medium. The processor of the computer device reads the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the violation audio detection method provided in the above-mentioned embodiments;

Violation audio detection methods, including:

Obtain the audio to be detected, extract the human voice audio in the audio to be detected, and convert the human voice audio into text information;

Text information is input into the illegal tone detection module to detect whether there is an illegal tone in the audio to be detected; wherein, the illegal tone detection module is based on at least one of the tone words in the text information and the duration of the audio to be detected;

If there is an illegal tone in the audio to be detected, the text information is input into a preset violation detection model to detect whether the audio to be detected is an illegal audio.

The above content is only a preferred exemplary embodiment of the application, and is not intended to limit the implementation of the application. Those skilled in the art can easily make corresponding modifications or modifications according to the main idea and spirit of the application. Therefore, the scope of protection of this application should be based on the scope of protection required by the claims.

Claims (10)

1. A method for detecting offensive audio, comprising:

acquiring audio to be detected, extracting voice audio in the audio to be detected, and converting the voice audio into text information;

inputting the text information into a violation voice detection module to detect whether the violation voice exists in the audio to be detected; the offensive word and phrase detection module detects at least one of word and phrase in the text information and duration of the audio to be detected;

if the voice frequency to be detected has the offensive voice, inputting the text information into a preset offensive detection model to detect whether the voice frequency to be detected is the offensive voice frequency.

2. The method of claim 1, wherein said inputting the text information into a offensive detection module detects whether offensive is present in the audio to be detected, comprising:

acquiring duration information of the audio to be detected, and extracting the number of words and words in the text information and the number of text words of the text information;

calculating a first ratio of the number of the Chinese words to the number of the words, and calculating a second ratio of the number of the text words to the duration information;

and detecting whether the offensive language gas exists in the audio to be detected according to the first ratio and the second ratio.

3. The method of claim 2, wherein detecting whether there is a offensive language in the audio to be detected based on the first ratio and the second ratio comprises:

matching the first ratio with a first detection threshold to obtain a first matching result, and matching the second ratio with a second detection threshold to obtain a second matching result;

if the first matching result represents that the first ratio is larger than the first detection threshold, the second matching result represents that the second ratio is smaller than the second detection threshold, and it is determined that illegal voice exists in the audio to be detected;

And if the first matching result represents that the first ratio is smaller than or equal to the first detection threshold or the second matching result represents that the second ratio is larger than or equal to the second detection threshold, determining that no offensive is present in the audio to be detected.

4. The method of claim 1, wherein prior to said entering the text information into the offensive detection module to detect whether there is an offensive in the audio to be detected, the method further comprises:

extracting voiceprint features in the voice frequency of the person to obtain target voiceprint features;

matching the target voiceprint features with the illegal voiceprint features in the voiceprint library to obtain feature matching results;

and if the characteristic matching result represents that the target voiceprint characteristic is not matched with the illegal voiceprint characteristic, executing the step of inputting the text information into a preset illegal voice gas detection module to detect whether illegal voice gas exists in the audio to be detected.

5. The method of claim 4, wherein the extracting voiceprint features in the human voice audio results in target voiceprint features, comprising:

carrying out segmentation processing on the voice audio to obtain a plurality of voice frequency bands;

Extracting voiceprint characteristics of each voice audio segment through a voiceprint model;

and calculating average voiceprint features of the plurality of voiceprint features, and taking the average voiceprint features as target voiceprint features.

6. The method of any one of claims 1 to 5, wherein the method further comprises:

if no offensive language exists in the audio to be detected, inputting the text information into a preset semantic model for semantic recognition processing to obtain a semantic recognition result;

and determining whether the audio to be detected is illegal audio according to the semantic recognition result.

7. The method of claim 6, wherein after said determining whether the audio to be detected is offending audio based on the semantic recognition result, the method further comprises:

if the audio to be detected is illegal audio, determining the type of the violation according to the semantic recognition result;

and determining a corresponding processing flow according to the violation category, and processing the audio to be detected through the processing flow.

8. An offensive audio detection device, comprising:

the acquisition module is configured to acquire audio to be detected, extract human voice audio in the audio to be detected and convert the human voice audio into text information;

The first detection module is configured to input the text information into the offensive language gas detection module to detect whether the offensive language gas exists in the audio to be detected; the offensive word and phrase detection module detects at least one of word and phrase in the text information and duration of the audio to be detected;

and the second detection module is configured to input the text information into a preset violation detection model to detect whether the audio to be detected is the violation audio if the violation air exists in the audio to be detected.

9. An electronic device, comprising:

one or more processors;

storage means for storing one or more programs that, when executed by the one or more processors, cause the electronic device to implement the method of detecting offending audio as claimed in any of claims 1 to 7.

10. A computer readable storage medium having stored thereon computer readable instructions which, when executed by a processor of a computer, cause the computer to perform the method of detecting offending audio of any of claims 1 to 7.