AVA - AUTONOMOUS VIRTUAL ASSISTANT

AVA - Autonomous Virtual Assistant is our state-of-the-art technology that can extract the necessary information from a small video/several pictures and a recording of a person’s face and afterwards it will generate an extensive video of the person while speaking a given text.

What is AVA?

We are currently working at the development of our own state-of-the-art technology, AVA - Autonomous Virtual Assistant. AVA is a unique technology that we are fully developing in-house. Its purpose is to become a comprehensive system that performs facial and/or voice transfer between multiple types of sources and references. Ava extracts detailed information from features such as facial motions and expressions, eyelid shapes and movement to create an accurate representation of the person. AVA is flexible as it allows multiple types of references: image or video, 3D avatar and voice. It offers high fidelity and a modular design.

AVA can be used as a

IoT digital assistant (displayable on various household items)
In-game character
Cartoon character
Chat app avatar

AVA's main features are:

Speech2Animation

Train Neural Network to learn lip syncing
Modelled as a sequence to sequence mapping from audio to video tailored for a specific individual
An LSTM RNN learns the mapping from raw audio features to time-varying sparse mouth shapes
Needs extensive training data

Face2Face Transfer

Architecture: CycleGAN
Purpose: generate the face image of the target character with the corresponding head pose and facial expression of the source.
Discriminator: 2 discriminator - PatchGAN
- 1 D with small receptive field (42x42) to encourage G to learn a local structure.
- 2 D with large receptive field (97x97) to encourage G to learn to generate realistic poses

Avatar Digitization

Real-time 3D Eyelids Tracking from Semantic Edges:
- A technique that reconstructs 3D shapes and motions of eyelids in real-time
- Generates full-face results with more detailed eye regions
- Applicable for different human races, eyelid shapes and motions

Technologies that are behind AVA:

Convolutional Neural Networks (CNN)
Generative Adversative Networks
Tensorflow (Python, C++)
Cloud based backend

Would you like to know more about this project? Reach out today!

We use cookies to improve your experience. By your continued use of this site you accept such use. Please see our privacy statement for more information about cookies. Learn more