PreventAI

PreventAI, a project of Swedish startup InfraMotionLabs, is an AI-powered surveillance system for supermarkets. I developed the client-facing platform prototype and an AI virtual assistant capable of human-like speech and behavior.

I developed the 3D model of the avatar using advanced modeling techniques in blender . To bring the avatar to life, I implemented a complex system of shape keys, which allowed for precise control over facial expressions, particularly the mouth and lip movements. This technique involves creating multiple variations of the mesh and blending between them to achieve smooth, realistic animations.

For the speech synthesis and lip-syncing, I integrated the azure Cognitive Services Text-to-Speech API. This powerful tool not only generates high-quality audio but also provides detailed viseme data. Visemes are the visual representations of phonemes, which are the smallest units of sound in speech. By mapping these visemes to the corresponding shape keys in the 3D model, I achieved a remarkably accurate synchronization between the avatar's mouth movements and the generated audio.

The result is a highly realistic AI virtual assistant capable of natural-looking speech, enhancing the user experience and bringing a human

As PreventAI is an emerging startup, I can only share limited information about the project. The video clip above showcases an early version of the AI avatar, demonstrating the capabilities we've developed so far.