Microsoft

Voice transformation occupies 2 of the 4 AI and machine learning patents, addressing how players sound in multiplayer role-playing games. The first system analyzes both player and character physical attributes through computer vision, then applies real-time voice transformation to automatically match voices to character appearance while reducing harassment potential. The second takes this further by adjusting vocal output frame-by-frame based on character state, making voices sound appropriately breathless during exertion or muffled underwater while maintaining the speaker's natural speech patterns. Beyond voice work, the category includes an avatar generation system that trains Gaussian splatting models with human appearance priors, allowing photorealistic digital twins to be reconstructed from limited input data. A fourth patent replaces traditional inverse kinematics with neural networks that understand hand biomechanics and object interaction, generating realistic hand poses both for real-time animation and as synthetic training data for robotics applications.

The single Audio patent extends precomputed wave-based acoustic modeling to handle doors and openings that change state during gameplay. Previous techniques required fully static environments, but this approach stores portal path data separately from ambient acoustic parameters, allowing games and VR experiences to recalculate sound propagation cheaply when a door opens or closes without recomputing the entire acoustic scene.