Nvidia Maxine

Main articles:

• Types and possibilities of the VIDEOCONFERENCING modern systems
• IT products and online services for remote work

Nvidia Maxine is a cloud platform of artificial intelligence for video streaming.

2020: Announcement of Nvidia Maxine

On October 5, 2020 the NVIDIA company announced the NVIDIA Maxine platform which provides to developers the cloud GPU accelerated software based on the artificial intelligence (AI) for video conferences created for optimization of a streaming video.

Nvidia Maxine

According to the company, the providers of a video conferencing using the platform on the graphic processors NVIDIA in a cloud can offer users opportunities of artificial intelligence, including correction of a look, the optimized permission, noise reduction, repeated illumination of the person and others.

As data are processed in a cloud, but not locally, end users can use these opportunities without special hardware.

For October, 2020 video conferences became a part of our life, helping people to work, study and play, and even to consult at doctors. Ian Buck, the vice president and the director of NVIDIA of the accelerated calculations speaks

The Maxine platform reduces the bandwidth required for video calls. Instead of stream transfer of all pixels of the screen the smart software analyzes key features of each person participating in a call and then intellectually recreates the person on the other hand. It allows to reduce the amount of data of a streaming video, sent on Network there and back.

Using this technology of video compression on the basis of AI working on the graphic processors NVIDIA, developers can lower load of bandwidth to one tenth from requirements of the compression standard of a streaming video of H.264.

Developments of researchers of NVIDIA which will be included in Maxine will make a video conferencing more similar to personal meeting. Service providers of a video conferencing will be able to use the researches NVIDIA in generative and competitive networks (GAN) to offer a set of functions.

For example, function of alignment of the person allows to align automatically a position of the person so that it seemed that during the conversation people face each other, and function of correction of a look helps to imitate visual contact even if the camera is not combined with the user's screen. As the volume of video conferences since the beginning of 2020 increased by 10 times, these functions help people to focus on a conversation, but not on the camera.

Developers can also add functions which allow participants of a call to select own animated avatars with the realistic animation which is automatically managed by their voice and emotional tone in real time. The option of an automatic frame allows a video flow to trace the one who speaks at present even if it moves away from the screen.

Using functions of dialogue AI based on SDK NVIDIA Jarvis, developers can integrate the virtual assistants using modern language models of AI for speech recognition, understanding of language and speech generation. Virtual assistants can take notes, set actions and answer questions with a human voice. Additional services of dialogue AI, such as transfers, subtitlings and a transcription, help participants to understand what is discussed during video conference.

According to the representative of the company, it is difficult to predict demand for a video conferencing at a certain point in time if hundreds or even thousands of users try to join a call. NVIDIA Maxine uses the AI microservices working in clusters of the containers Kubernetes on the graphic processors NVIDIA to help developers to scale the services according to current demands. Users can start several AI functions at the same time, without exceeding requirements of applications for delays.

Service providers of video conferences can use Maxine to give AI opportunities to hundreds of thousands of users, executing inferens on the graphic processors NVIDIA in a cloud. Modular construction of the Maxine platform allows developers to select possibilities of AI for integration into the solutions for a video conferencing.

The Maxine platform integrates technologies from several SDK NVIDIA and API. In addition to NVIDIA Jarvis, the Maxine platform also uses SDK NVIDIA DeepStream for high-speed audio streaming and video and SDK NVIDIA TensorRTTM for a productive inferens.

Developers of AI-applications of computer vision, partners in the software, startups and the computer makers creating audio-both video-applications and services for October, 2020 can submit the application for early access to the NVIDIA Maxine platform.