Usually, these effects are recorded by Foley artists in the studio who use specific objects to produce special sounds. For example, if the director wants to add the sound of broken glass to the video, he may need to break many pieces of glass by Foley artists until the sound matches the video clip. (Note: Foley, or onomatopoeia, is used to enhance the auditory experience of movies. It is a reproduction of daily sound effects. It will be added to movies, videos and other media in post production to improve audio quality. Named after sound artist Jack Foley, these reproduced sounds can range from the sound of clothes and feet to creaking doors and broken glass.) An early Foley The artist (second from right) adds sound effects to a radio program on the spot, and his palette can simulate the effect of a phone ringing and closing a door (source: Wikipedia) < / P > < p > to solve this problem, researchers from the University of Texas at San Antonio have created an automatic program based on machine learning, which can be used to identify a video clip What’s going on, and creating realistic sound effects based on it, has convinced most people that fake sound effects are real. < / P > < p > related research results were published in the IEEE multimedia transactions (< / P > < p > in the form of papers published by the IEEE Computer Association, IEEE Communications Association and other scientific journals Since the s, it has been a complicated thing to add sound effects to specific scenes by using Foley art in the post production of movies and TV series. If there is no realistic Foley sound effect, the film will appear empty and unreal, but the synthesis of Foley sound effect increases the creation cycle and cost of film and TV series. ” < / P > < p > in the multi-layer machine learning program created by Prevost and his doctoral student sanchita ghose, they created two different models respectively for the step of recognizing the action in the video and determining the appropriate sound to test the synthesis effect of sound effect. < / P > < p > the results show that their model is most suitable for scenes where the time does not need to be exactly the same as the video (for example, the sound of rain and the crackle of fire), but does not perform well when the video contains random actions (such as typing, thunderstorms) over time. In addition, they also surveyed 57 local college students to see if they could tell which movie clips contained the original sound effects. When evaluating the first model, 73% of the students chose the composite sound as the original sound effect, and 66% when evaluating the second model. It can be seen that these two AI models have deceived most people.