LipConfirm
The next generation of autocorrect.
Autocorrect that uses your lips as a missing signal of intent.
Typing on small portrait phone keyboards remains a major source of errors and frustration. LipConfirm fuses keystrokes with real-time lip cues (and optional whisper-level audio) to meaningfully improve the accuracy of autocorrect.
Silent, easy input
Combines subtle lip cues with normal typing to boost accuracy. No need to talk out loud. Many people already mouth words naturally while they type.
A missing intent signal
Adds a missing signal layer that helps resolve high-ambiguity or messy typing cases that even leading mobile keyboards still struggle to interpret.
Why Now?
Even LLM-informed autocorrect still struggles when portrait-mode mobile typing produces garbled or ambiguous words. LipConfirm adds a new private signal — silent lip movement — to help recover what the user meant to type.
How it works
Capture
Keystrokes and opt-in mouth-region cues are captured locally during typing, with privacy built in. The camera and microphone stay off unless enabled; optional whisper-level audio can be used as an extra signal. Raw frames and audio are processed on-device and discarded after feature extraction or scoring.
Multimodal fusion
LipConfirm combines typed input with lip cues and optional whisper-level audio to improve correction ranking and candidate recovery. The architecture can use heuristic scoring, on-device ML, or learned multimodal fusion depending on the implementation.
Seamless correction
When confidence clears a threshold, LipConfirm can replace the garble without interrupting your flow. In lower-confidence cases, the normal suggestion bar remains available as a fallback.
What the demo shows
- Messy input, clear intent. Fast, imperfect typing that normally fails to suggest the right word.
- Lip-guided selection. As the user silently mouths the intended word, LipConfirm combines mouth-region cues with the typed string to push the intended correction toward the right choice.
- Typing stays primary. This is not voice dictation and not standalone lipreading. The user keeps typing normally; lip movement acts as an auxiliary intent signal. Whisper-level audio can optionally add a second private signal.
- Note: The lip video panel is shown here as a behind-the-scenes visual aid. In normal use, processed lip cues are not displayed on screen.
What the demo shows
- Messy input, clear intent. Fast, imperfect typing that normally fails to suggest the right word.
- Lip-guided selection. As the user silently mouths the intended word, LipConfirm combines mouth-region cues with the typed string to push the intended correction toward the right choice.
- Typing stays primary. This is not voice dictation and not standalone lipreading. The user keeps typing normally; lip movement acts as an auxiliary intent signal. Whisper-level audio can optionally add a second private signal.
- Note: The lip video panel is shown here as a behind-the-scenes visual aid. In normal use, processed lip cues are not displayed on screen.
Strategic Partners & Investors
From desktop proof-of-concept to mobile prototype
LipConfirm has an early desktop proof-of-concept demonstrating lip-guided autocorrect reranking. The next milestone is completing our mobile prototype, running real-user tests, and optimizing for private on-device operation.
Built for existing input stacks
LipConfirm is designed as an auxiliary correction layer that can work with existing keyboard, autocorrect, spell-checking, and language-model infrastructure, rather than requiring a full rewrite of the text-entry stack.
High-value use cases
LipConfirm is especially relevant to mobile keyboards, accessibility, quiet or private input, wearables, smart glasses, and AR/XR text entry, where small keyboards and limited feedback make correction harder.
LipConfirm is open to strategic licensing, integration, SDK partnership, and investment discussions with companies and investors focused on next-generation human-computer input. A short non-confidential overview and prototype demo are available for serious strategic or investment inquiries.
Patent & IP
Patent pending. LipConfirm is the subject of a pending U.S. nonprovisional utility patent application, with Track One prioritized examination granted by the USPTO. The application is directed to multimodal keyboard/autocorrect technology that uses lip-movement cues, and optionally whisper-level audio, as auxiliary intent signals to improve correction accuracy while preserving the ordinary typing workflow.
FAQ
Is this just voice dictation?
No. LipConfirm works silently from lip cues (visemes) fused with keystrokes, without requiring audible speech. Whisper input is optional, not required.
Does this require the internet or cloud processing?
No. All processing runs fully on-device in real time. No frames or audio are stored or transmitted.
Will this drain battery or slow down typing?
LipConfirm is designed for lightweight, on-device operation and for activation only during relevant typing intervals. Final battery impact will depend on the implementation and device settings.
Is my camera always on?
No. The camera activates only during typing sessions, and only with user consent. Frames are analyzed instantly and discarded.
How is this different from existing autocorrect?
Existing autocorrect primarily relies on typed characters, keyboard geometry, context, and language-model predictions. LipConfirm adds another signal: mouth-region movement that can help identify what the user intended to type, especially when typed input is garbled or ambiguous.
Do I need to perfectly mouth the words?
No. The system captures multiple cues from your lip movements — like syllable counts and key phoneme shapes — to help disambiguate your intended word alongside your typed input. Lip input is natural and easy to get used to, and it’s far more convenient and faster than having to backtrack and retype on a mobile keyboard.
Can I just use lip cues without typing?
The first product is typing-first: LipConfirm is most powerful when typed input remains the primary signal and lip cues resolve ambiguity. In specialized modes, lip-only or whisper-assisted operation may also be possible, but the core value is fusion with ordinary keyboard input.
Why is whisper audio part of this?
Whisper input is optional and exists for maximum flexibility across environments. In low-light situations, lips may not be visible but a quiet whisper can be captured. In very noisy places, whispers may not work but lip cues still do. Lip visemes remain the primary signal, but whisper audio provides an optional extra layer of robustness across different environments.
How is AI used here?
LipConfirm can use on-device ML models to combine typed input with lip cues and optional whisper-level audio. Existing autocorrect primarily relies on typed characters, keyboard geometry, context, and language-model predictions. LipConfirm adds mouth-region movement as an additional intent signal that can help rerank candidates, recover missed candidates, or improve confidence in the intended correction.
Does the user need to train or calibrate the model?
LipConfirm is designed to minimize or avoid required calibration in ordinary use, with optional personalization available to improve accuracy over time.
How much accuracy improvement does LipConfirm provide?
LipConfirm is designed to recover missed cases where ordinary autocorrect fails because the intended word is ambiguous, low-ranked, or absent from the suggestion list. Early prototype and modeling examples indicate that lip cues can improve recovery in challenging inputs, but real-world accuracy will depend on implementation, device conditions, and user behavior.
Is LipConfirm only for mobile phones?
Mobile keyboards are the primary target because fast typing on small screens, especially in portrait mode, causes the most frustration today. But LipConfirm is designed as a general fusion layer that can work in any typing environment — including tablets, laptops, and even wearable devices.
Will I still see suggestion lists?
Less often. When the fusion signal is strong, LipConfirm auto-commits the intended word so you don’t have to pick from a list. If confidence is lower, the normal suggestion bar appears as a fallback, but enhanced with more accurate choices compared to keyboard input alone.
Is this patented?
LipConfirm is patent pending. A U.S. nonprovisional utility patent application has been filed, and Track One prioritized examination has been granted by the USPTO.
Contact
Email: hello@lipconfirm.com
LipConfirm was conceived and developed by Andre Persidsky, a multidisciplinary engineer and inventor with experience building, patenting, and licensing technology. Andre has previously licensed technology to commercial partners including Spotify, and has led hardware/software health-technology projects in collaboration with external research institutions.