Visper Visper Virtual Character Creation Platform

Main articles:

• Speech technology: from recognition to understanding
• Virtual Assistant Creation Platforms

2021

Platform availability in English

On November 3, 2021, Sberbank announced that the Visper platform, which allows you to generate videos with photorealistic and animated characters, became available in English.

visual content Users around the world can now create with it. In addition to the translation of the interface, the option of generating speech in English appeared: just enter the text and the avatar reads it with the correct pronunciation.

If earlier Visper users could create a video understandable to 200 million people who speak Russian, now the potential audience has increased by more than 8 times - thanks to about 1.5 billion people who speak English.

Another characters also appeared on the platform: you can now make videos with three realistic avatars (Elena, Peter and Eve) and three animated ones (Matilda, Vi and Kira). Various images and costumes are available for them, 4 female and 2 male voices for Russian voice acting and 5 female and male voices for English. You can download your own record - and then the facial expressions of the character will adjust to it. Each avatar's speech can be marked intonationally: put pauses and stresses, escort with gestures, put music and background. The video can be based on any presentation, as well as animate your own photo and insert it into the video.

Visper allows content makers and information product developers to create high-quality materials faster and cheaper and experiment with formats. With this platform, you can quickly create creative videos without expensive and long shots, animate text content using videos, quickly receive visual material for dynamically changing content (news, announcements) and personalize communication.

Every month, any user can create two minutes of video for free. Both per-minute payment and subscription are available, and for business - and personal tariffs based on the specifics of the activity

"The platform was launched only six months ago, and it is already available to users around the world. At the beginning of November 2021, 17 thousand people use the service, who daily create videos with our characters, each of which is suitable for different audiences and purposes. The creation of an English-language interface made it logical to continue the development of the project. This will allow us to attract an international audience and introduce the world to our developments, " says David Rafalovsky, executive vice president of Sberbank, CTO, "" Sbera head of the Technology block.

Animated avatar Matilda

On July 15, 2021, Sberbank announced that the Visper platform had opened up updated opportunities for its users to create videos and presentations. Two images appeared in the update - the first female animated character Matilda on the platform and the realistic male character Peter. Also, without special tools, you can now place pauses and stresses in the speech of avatars and download your own audio file with text for which the facial expressions of the character are adjusted. And you can animate the video by downloading background music.

Matilda

According to the company, the platform has the opportunity to create videos with animated avatars, the first of which was Matilda, developed by the Visper team together with ARVR Sber/Lab. When using this image, all the same functions as realistic characters are already available: reading out text on, in Russian adding your own audio track, choosing a voice model (neutral, friendly or business). Also, a male character was added to five female realistic images, for which neutral and kind sound models are available.

One of the updates was the first animated character on the platform. Now it is easier for users without special skills to make beautiful and memorable videos and presentations in different styles that can be used for business. said Konstantin Kruglov, CEO SberDevices, senior vice president for digital surfaces at Sberbank

Another update was the text markup function on the stage, which the character reads out. Now you can make the speech of the avatar more emotional, changing its pace, adding pauses and stresses. This is done very simply: you only need to put the appropriate marks in the text - and the speech of the avatar will sound different.

In addition to the ability to control speech synthesis, there is an option to download your audio file with text that the character will voice. The program itself analyzes the record and synchronizes facial expressions to make it look as natural as possible. In user records, there are often foreign noises that are automatically suppressed, which optimizes the work on the roller and the quality of the generated material. This function also makes it possible to use speaker speech and recordings in other languages.

Also, now in the video you can even download background music, which will be played throughout the video at a volume of 10% of the sound of the character's text. In order to be able to quickly try how it works, the platform for July 2021 has three tracks created using deep learning neural networks developed in SberDevices.

Launch Platform

On April 26, 2021, Sberbank launched the Visper virtual character platform.

As of April 2021, using the service, any user can animate the presentation or create a video with visual characters. Characters can be used to solve problems in various areas: in marketing, for online training services, creating entertainment and news media content, as well as content for social networks. However, soon the platform's capabilities will expand significantly, including replenished with solutions for gamedays of the industry, when on the basis of Visper tools developers will be able to create avatars and characters for games. During the beta phase, access to the platform will be free.

The video content market is huge, and almost all business segments sooner or later require video formats to promote their services. Visper developers have proposed a semi-automated process of turning a presentation into a video with visual characters, making it simple and not requiring technical skills. You just need to set the desired parameters, then the platform will do everything itself. Users will receive a tool for creating videos from pre-created presentations of classic formats (for example, PDF).

By creating a virtual character, the user can:

• Choose the appearance of the character - various options for the image and style are available;
• choose a voice model for character speech synthesis - neutral, friendly or business;
• add gesticulation, for example, gestures of appeal or likes - the character can say hello to the audience or turn to them during the story;
• Add an image or video as the accompanying material for the character video.

The platform's web interface is easy to use, and as of April 2021 has two main sections: desktop and desktop.

All previously generated videos, as well as incomplete drafts, are collected on the desktop. The transition to video creation is carried out in three versions - from scratch, from ready-made presentations in PDF format, or based on previously generated videos. In addition, by default, the user gets access to three template documents, from which he can create his own videos.

In the section for editing, you configure the character and add text for the movie. Here you can adjust the location and scale of the character and content in the frame, turn off the content or visual display of the announcer, change the backgrounds, as well as listen to fragments of synthesized speech and the entire scene. From the Stage, you can send the movie to generation or save it for editing.

Before creating any video, the user enters the virtual character control page, where he selects the background, voice and image, and also gives the character a name. It also displays previously created characters and there is an option to generate new ones.

One of the main features of Visper characters is the presence of gesticulation, which, as in life, helps to focus on key theses and keep the attention of the audience. In the near future, the option of automatic selection of gestures will be added: the user will have enough to download the text and check whether the visual character received the correct hints from the algorithms. Moreover, very soon the platform will be replenished with tools for generating three-dimensional characters.