Sony filed 14 patent applications across 7 categories: AI & Machine Learning (5), Audio (2), Cloud Gaming (2), Graphics (2), VR & AR (1), Streaming (1), and UI/UX (1).
The AI & Machine Learning patents cover speech synthesis systems for character dialogue, spectral rendering optimization, personalized trophy designs, and automated gameplay assistance. Audio technologies include dynamic prioritization of game sounds and communications, plus personalized sound modification based on player preferences. Cloud_gaming applications describe predictive frame generation during network interruptions and split-screen configurations that combine local and remote gameplay, while Graphics patents detail ray-tracing optimization and neural network-based 3D object rendering.
Two Audio patents tackle the challenge of managing competing sound sources during gameplay. The first uses AI to identify which game moments and communications matter most, then automatically adjusts volume levels across in-game Audio and external streams to prevent critical information from being masked. The second learns individual player preferences over time and modifies specific in-game sounds to reduce listening fatigue, muting or replacing Audio elements while preserving sounds the developer marks as essential to the experience.
Cloud gaming gets attention through 2 patents focused on network reliability and resource distribution. When packets drop during Streaming, one system generates probable next frames on the client device rather than waiting for the server to resend missing data, maintaining visual continuity during connection hiccups. The other enables split-screen multiplayer sessions where one player's game runs on the local device while a second player's experience streams from a cloud server or remote console, allowing devices with limited processing power to host multi-player sessions.
A single VR & AR patent addresses privacy concerns in head-mounted displays by filtering motion sensor data to remove voice-related vibrations while preserving the tracking information needed for immersive experiences. The filtering can be tuned based on the application and user preferences, giving players and Platforms operators control over what information the sensors capture.
Five AI & Machine Learning patents span character dialogue, personalization, and player assistance. Two relate to text-to-speech systems for game characters, with one allowing pitch adjustments at the phoneme level through quantized bins and convolutional networks, while the other separates pitch prediction from pitch application to give developers manual control over emotional delivery without re-recording voice actors. A third generates personalized trophy designs based on player behavior and gameplay context, creating achievement presentations tailored to how someone plays rather than showing identical rewards to everyone. The fourth analyzes gameplay video frames to predict what questions a player might have and proactively generates answers without waiting for manual queries. The fifth automates the creation of transitional content for highlight reels, using multiple AI components in a feedback loop where one generates filler material, another evaluates quality, and a third refines the process until the output meets specified thresholds.
The Streaming patent converts flat 2D gameplay video into 3D reconstructions that spectators can explore from different camera angles, giving viewers control over their perspective independent of the player's viewpoint during live broadcasts.
UI/UX receives coverage through 1 patent that determines which non-player character someone intends to interact with by analyzing gaze direction and proximity, solving the common problem of accidentally engaging the wrong character when multiple NPCs occupy the same space.
Graphics rendering appears in 2 patents that both leverage AI to improve visual quality while reducing computational demands. One distributes spectral ray-tracing calculations across neighboring pixels and successive frames rather than computing multiple wavelengths for every pixel in every frame, selecting wavelengths intelligently based on material properties. The other integrates neural radiance field techniques into traditional mesh rendering pipelines, letting the neural network handle only surface details while the mesh manages geometry and motion, achieving photorealistic results faster than full neural rendering approaches.
All data sourced from USPTO patent filings. Google Patents may take several weeks to index recent publications. If a link is unavailable, search for the patent number at USPTO Patent Public Search.