Destiny!What should I do?

Original: Lee Di

In the new media era, people are keen to brush video news and watch live broadcast. However, if the audits and live broadcasts are not strict, wrong materials, and information are wrong, the media's authority and credibility will withstand the impact of index levels.

Where is the pain points of video and live audits? How can we improve the security of video and live broadcast? Are there any reliable AI audits in the industry available?

With these questions, the media tea will talk to a media technical person in charge to observe the video audit industry up close.

Where is the news and video and live broadcast review?

With the development of the video industry, the media's review of the content of video and live broadcasts has become more and more important.

However, when there is always a intention to lose Jingzhou, the audit process is taken, and the news sent out either the officials in the video, or the picture appears bad artists, or the material infringement used, or the geographical coordinates that should not be exposed during the live broadcast, shooting it, shooting, shooting, shooting. On the incorrect national flag, national emblem, map and other contents, or in the live broadcast of incorrect national flags, national emblems, maps, etc.

In comparison, the process of reviewing the "Three Trials and Three Schools" process can also strictly control the risk of content, and errors in text and pictures can be modified or deleted within a valid time.

The picture comes from the petal Meisu

However, the difficulty of video and live audits is too difficult. Where is the difficulty?

On the one hand, there are many videos and live broadcasts, the information density is high, and the information scenarios presented are complicated. For example, the live broadcast is performed in real time. The "Statistical Report of the Development of China's Internet" shows that as of December 2021, the number of online video users in my country reached 975 million, an increase of 47.94 million compared with December 2020, accounting for 94.5%of the overall netizens. This means that the demand for videos and live broadcasts will be greater and the communication will be wider, and the accuracy and speed of the audit will be higher. Once a problem occurs, it is difficult to withdraw and modify, which is extremely damaged to the media.

In addition, at present, the media industry's internal video audit is still mainly manual review. Due to the richness, density and the differences of editing knowledge structure, it is difficult to ensure that each editor can make errors in the content of the video in the video. All accurately identify, causing many important videos to watch it collectively. A few minutes of video may take a few hours to perform manual review during the review stage, which is time -consuming and laborious.

At the same time, in recent years, in response to the issues of video and live broadcasts, the relevant regulatory authorities have increased the supervision of video and live broadcast and improved the review requirements. "" "The General Principles of Internet Audiovisual Programs" and other documents have detailed strict regulations for video production, review and release. As an industry organization, the China Internet Audiovisual Program Service Association also released the "Detailed Rules for the Review of the Network Content Review (2021)", which further detailed the review standard for the industry.

There are regulatory requirements, the pain points inside, and the industry calls the reliable video review helper.

Realize no differential review, AI strong complementary manual review

For the media, what kind of review helper is needed?

A person in charge of a media technology said that at present, the mainstream media is looking forward to a video audit system that can realize the full scene coverage and full factor review. For example, it can be differentiated with multiple scenarios such as long and short videos and live broadcasts; there is an operating interface that is easy to get started, which can accurately review the characters, backgrounds, sound patterns, emotions, text and other content in the video. Video technical trials (including screen jitter, flower screen, low light, overtime, etc., no audio rails, mute, low volume, blasting sound, etc.) can also be completed through video audit software.

Under this standard, the first AI video audit is to be able to review frameworks, and each frame cannot be leaked. This is one of the core standards; the second is to review whether the video content meets the mainstream values ​​and accurately identify The content that does not meet the mainstream values; the third is the full coverage review of the full factor (characters, scenes, sounds, OCR, etc.) of the video.

In response to demand, the video industry is also constantly developing AI audit technology, and has gradually been introduced into business processes by mainstream media, such as "Net Qing content security brain" used by many central media.

The advantage of AI audit is that there is no differential review of all review elements, which can make up for the abundance of manual review.

The so -called non -difference review is relative to manual review. When watching video content, it usually depends on the core elements, such as the characters, sounds, and eye -catching logo in the video. For some background content, such as the background people and the background scene, they may not be able to see it so detailed. For AI review, these elements are consistent with the main element review accuracy, without differences.

Therefore, it is necessary to introduce AI audit technology. On the one hand, AI audits can systematically split and filter the content to reduce the burden on manual review.

Taking the content of the content of the content of the net, the video audit function can split the video content for structure, including screens, audio, OCR, sound pattern, emotion, technology and other latitude. At the same time, the types of problem types are subdivided, and a variety of issues such as politics, terrorism, pornography, bad values, advertising content recognition, technical audit, and technical review can be effectively reviewed. The accuracy and accuracy rate can reach more than 99%.

Its live audit function will conduct real -time review during the live broadcast. Without affecting the recognition rate, not only reviews the live broadcast screen, such as the recognition prompts of minor content, but also include the barrage of the rapid jumping out of the live broadcast, commenting , Filter sensitive content such as politics, yellow, and violent terror, and then push the results of the machine review to a special person, conduct one -on -one or one -to -many concentrated review, so that the results of the machine review will quickly assist artificial review. At the same time, the audit system supports docking mainstream live broadcast platforms, and can return to the audit recommendation and associated conductors through API in real time. Net Qing content security Brain conducts supervision and identification and marking reminder of minors appearing in the live broadcast

On the other hand, AI audits can also give analysis on the identified issues, effectively complement each other with manual audit, and greatly improve the efficiency of video audit.

For example, when using past materials, some editors may not understand the information and backgrounds of video characters, which leads to some officials, reactionary split characters, and inferior artists appear in the video. In response to this kind of problems, the content of the content of the content of Net Qing can conduct effective checks according to the model, remind the editor to review and check in time, and make treatment.

Net Qing content security The brain identifies and labels the officials of the horse

Review the patent frame by frame to help the media take a good review

Nowadays, there are many AI audit products on the market, and the effects are different. How to choose suitable for you need to fully understand the advantages of AI audit products. Compared with similar products, the video audit of Net Qing content security brain has the following advantages:

First, its core advantage lies in frame -by -frame review technology -real -time video content dynamic prediction identification algorithm.

This algorithm can completely throw away the restrictions on the fixed timing of pixels (frame clock) control through the principle of human eye visual retina and brain working principle, and control the change of the signal amplitude domain. %.

Net Qing content security Brain uses frame -by -frame review technology to accurately position and identify violent terror guns

For example, when a gun appears in the video screen, the system will mark and prompt the gun during frame -by -frame review to facilitate editing to deal with them in time.

Compared with the traditional frame method, the advantage of the algorithm is that it can make the total amount of the overall data required in the video less than 100 times, and overcomes the problem of omissions of the content information of the frame between the frame of the traditional frame. Internal recognition accuracy.

Second, the model is perfect. Net Qing content security The brain can cover the facial detection model, the character detection model of political affairs, the flag detection model, the sound pattern identification ability, the logo detection model, the inferior artist detection, the two -dimensional code/barcode recognition, and OCR text recognition Models can effectively identify error content.

When the national flag, national emblem and other political logos appear in the video scenario and national emblem, the audit system will make corresponding prompts in the video according to the detection model and flag detection model of the government involved in politics, so that the editorial review is convenient to avoid the details due to details. Step on the pit.

Net Qing content security The brain's prompt on the national flag

Net Qing content security The brain's prompt to the national emblem

Third, the split is fine and the feedback is rapid. Net Qing content security brain will split the video into the 6 major items of screens, audio, OCR, sound pattern, emotion, and technology, and further subdivide Show in the audit page.

For example, for the audio content in the video, the system will use ASR to identify and extract the voice content, conduct text review of the proposed content, and identify the sensitive points. And combined with sound pattern recognition to identify the speakers in the scene, as shown in the video demonstration case below, the real sound pattern of the characters can be recognized, the content of the speaker, and whether the speaker corresponds to the video screen. The video audit interface can "see" the detection content in the voice.

This is the correct case recognition demonstration, and the content is correct

At the same time, the test feedback is fast. For example, for text detection, the system can return the result within the millisecond level (1000 characters). Video detection duration: The total time of the video is long, and the result is returned within 1: 5 time. The results are detected within 0.1 seconds. The detection and detection rates of the contents of politics, yellow, terror, advertising, and bad values ​​are up to 99%.

Fourth, provide video technical trial. Technical problems in the video, for example, when video or live broadcast blurred pictures, mute, bass, blasting sound, black screen and other video technical problems, often cause headaches for the media. In response to these issues, the safety of the content of the content of Net Qing can quickly identify and position without omissions.

Net Qing content security brain conducts technical trials on the video, prompting the picture to be blurred

Fifth, the system interface is simple and strong. Through the content of the content of the content of the net, the editor can locate the problem frame without repeated watching the video when reviewing the video screen. Find the text in the picture and then review it, etc., greatly improving the review efficiency.

The accuracy of the audit is high, intelligent and efficient, and naturally favored by users. Moreover, for the media, improving the accuracy and security of the content is the top priority. Compared with your own money to develop AI audit technology, a reliable and effective third -party AI audit may be a good choice.

- END -