A group of researchers from the NTU School of Computer Science and Engineering have developed a new artificial intelligence (AI) program that they have dubbed “DIverse yet Realistic Facial Animations” (DIRFA). This is a step in the direction of combining reality with simulation.
This software takes an audio clip and a still shot of a person’s face and converts them into a three-dimensional video that properly portrays the facial expressions and head motions of the person in the audio clip.
Realistic Talking AI Faces
According to the research team, DIRFA has undergone thorough training on over one million audiovisual clips from more than 6,000 individuals that were gathered from an open-source database. This is in contrast to existing techniques, which have difficulty dealing with fluctuations instance and maintaining emotional control.
The objective was to anticipate cues from the speaker’s words and to coordinate them naturally with the actor’s facial expressions and head movement. The end product is a piece of software that facilitates the production of videos with an increased sense of realism.
The corresponding author of the study and the person in charge of it, Associate Professor Lu Shijian, stressed the potential impact that DIRFA may have on multimedia communication. He claimed that the application, which combines several methods such as AI and machine learning, has the potential to completely transform this industry.
“Our program also builds on previous studies and represents an advancement in the technology, as videos created with our program are complete with accurate lip movements, vivid facial expressions and natural head poses, using only their audio recordings and static images,” the principal investigator stated in a prepared statement.
Dr. Wu Rongliang, who received his Ph.D. from the SCSE at NTU, was the initial author of the study and brought attention to the complexities of speech variations as well as the quantity of information that they transmit beyond the linguistic content. He referred to their method as a pioneering effort in the field of artificial intelligence and machine learning, where it aims to improve performance in audio representation learning.
Potential Applications(New AI Program Creates Realistic)
The researchers think that the uses of DIRFA could extend across a wide variety of businesses and fields, including the healthcare sector. The potential exists for the initiative to greatly improve user experiences by enabling virtual assistants and chatbots that are more technically advanced and realistic.
In addition, DIRFA has the potential to develop into a helpful instrument for people who have difficulties with speech or facial expressions, making it easier for these people to communicate through the use of expressive avatars or digital representations.
Associate Professor Lu underlined the importance of making additional enhancements to DIRFA’s interface in order to grant users increased control over certain outputs. The team has stated that it is dedicated to improving the program’s features and expanding its capabilities despite the fact that it can generate talking faces with exact lip movements, expressive facial expressions, and genuine head poses.
The NTU researchers intend, in the not-too-distant future, to fine-tune DIRFA’s facial expressions by utilizing a wider variety of datasets that comprise a greater variety of facial expressions and spoken audio snippets.
The research article, which was published in the journal Pattern Recognition under the title “Audio-driven talking face generation with diverse yet realistic facial animations,” may be found here.