In plain words
Language AI mostly lives in the cloud: no internet, no intelligence. This patent covers a single device that performs real-time natural language processing entirely offline — text segmentation, topic and subtopic detection, text summarisation and query-based information retrieval — with no internet connection at any step.
The claimed system is deliberately format-agnostic at both ends: it accepts text, audio, video and images (live or static) as input, and returns its output in whichever form the user requests — text, audio, video or images. You can query it by voice or text with keywords, and it can hand results to remote devices over multiple network protocols.
Filed in 2019, well before on-device AI became an industry slogan, it targets exactly the situations where language help matters most and connectivity is worst: rural regions, travel, emergencies and privacy-sensitive work.
How it works
- Take anything inMultiple input devices feed the system — live or static text, audio, video and images.
- Process on the deviceSegmentation, topic and subtopic detection, and summarisation run locally in real time, using techniques such as tagging and lexical chains — no server involved.
- Ask it questionsThe user queries by keywords, typed or spoken, and optionally interacts to steer the result.
- Answer in any formatThe summarised output is returned as text, audio, video or image — the user's choice — and can be sent onward to remote devices.
Why it matters
Offline, private, on-device language processing is now the direction the whole industry is moving — phones ship with local models precisely for privacy and reach. Holding a granted 2019 patent on an end-to-end offline NLP device reflects an early bet that aged well.