LipVoice

Author Mohammed Al-MrehTotal Visits 20Visits in 7 Days 0

// preview

// description

Reads your lips through the camera and speaks trained phrases aloud

LipVoice is a browser-based communication aid for people who cannot easily speak. It watches your mouth through the camera, matches what it sees against phrases you trained yourself, shows its best guess, and can speak the result out loud. You can train phrases from several head angles with an animated pronunciation guide, add a starter set of essentials, upload a voice recording so playback uses that voice, and adjust automatic speech, strictness and pause length in Settings. The interface supports English and Arabic, and video is processed on the device rather than uploaded.

Highlights

Camera lip reading

Reads your mouth

Custom phrases

Train your own words

My voice cloning

Speaks in your voice

Bilingual interface

English and Arabic

How to Use

Turn on the camera
Open the site and click Turn on camera on the intro card, then allow camera access. Fit your whole face in the frame with your mouth clearly visible. The page states that video stays on your device and nothing is uploaded, so the reading runs locally in the browser.
Train your phrases
Open Train phrases. Type a phrase, watch the animated mouth guide and use Play, Pause, Replay and Slow to study the motion, then press Start recording and repeat the phrase while following the head-angle chips for Front, Left, Right, Closer and Farther so LipVoice learns it from more than one view. Load essentials adds a starter set if you prefer not to type your own.
Read and speak
Mouth one of your trained phrases and then pause. The What we think you said card shows the best guess, adds it to the Conversation list, and can speak it aloud automatically. Use That is correct, Say it again, Record a sample or Clear to correct or improve the result.
Personalise the voice
Open Settings to turn automatic speaking on or off and adjust the strictness and pause-before-predicting sliders. Open My voice to upload a short recording of your own or a patient's voice so later predictions are spoken in that voice, and use the test text field to hear a sample before relying on it.