Sber’s Visper virtual character platform greets Matilda, its first animated avatar

Jul 15, 2021

Major updates include speech synthesis management and facial expression animation using your own audio recording.

Visper, Sber’s platform designed for visual content generation with the help of digital characters, is offering users new possibilities when creating video clips and presentations to make them truly unique. The update introduced two new characters – Matilda, the first female animated character on the platform, and Pyotr, a realistic male character. You can also insert pauses and emphasis in the avatars’ speech and upload your own audio with text, which will be used to adjust the characters’ facial expressions. Video clips can be brought alive by inserting background music.

New characters

You can now create video clips with animated avatars on the platform. Matilda, created by the Visper team and Sber AR/VR Lab, is the first of these avatars. The same functions available for the realistic characters are available for animated avatars: text reading in Russian, adding your own audio tracks, and selecting voice models (neutral, friendly, or professional). A male character, equipped with neutral and friendly voice models, has been added to the five realistic female characters.

At SberDevices, we always strive to make the products we design to help people solve everyday tasks more convenient and effective. The update for the Visper platform serves that exact purpose. The platform’s first animated character is one of the new updates. Now users without specialized knowledge will find it even easier to make beautiful, memorable video clips and presentations in various styles that can be used for business purposes, too. We are constantly adding characters and features, so that the content created on our platform is high quality and truly unique.

Konstantin Kruglov

CEO, SberDevices; Senior Vice President for New Digital Surfaces, Sberbank

Emotions, music, and animating your own soundtrack

Another one of the new updates is a markup feature for the editing table, where the character reads the text out. You can now make your avatar’s speech even more emotional by changing its pace and adding pauses and emphasis. The process is very simple: mark the appropriate moments in the text and completely change the way the character’s speech sounds.

In addition to speech synthesis management features, you can now also upload your own audio file with text for the character to read out. The program will analyze the recording and synchronize the facial expressions so that they look as natural as possible. Background noise from user recordings is automatically suppressed, which significantly simplifies clip processing and improves the quality of the material generated. This function also allows clients to use speech and recordings in different languages.

Background music can now be inserted in a video. The music will play for the duration of the clip at 10% of the volume of the character’s spoken text. This feature helps users set the mood and make their video unique, livelier, and more dynamic. Users can try the feature out – three tracks created with deep learning neural networks designed by SberDevices have already been uploaded to the platform.

Further information about the Visper update in the video.