Accessibility-first communicationFoundation build · Phase 1

Sign.Speak.Stay safe.

SanketSetu is a browser-based communication workspace for Deaf and hard-of-hearing people, built around supported gestures, live captions, sound awareness, and quick phrases.

Every channel is optional. Every start is yours.

  • No account needed
  • Camera and microphone only on request
  • Preferences stay on this device
SanketSetu symbol: a bridge with signal rays

The SanketSetu mark

संकेत सेतु

A bridge between people.

  • Sign to VoiceExperimental
  • Voice to TextLive
  • Sound AlertsExperimental
  • Quick CardsLive
01The gap

Sound moves fast.
Access should keep up.

A great deal of daily information arrives as sound. For Deaf and hard-of-hearing people, that information often has no second channel.

SanketSetu is being built to add visual and expressive alternatives: a phrase you can show, captions you can read, a visual cue for a selected sound, and a small set of supported gestures.

It is not an attempt to replace interpreters, captioning services, hearing loops, or emergency systems. It is a companion for the moments when none of those are within reach.

Information that arrives only as sound has no second channel.

The design premise behind SanketSetu.

Ordinary moments, one channel only

  • A smoke alarm sounds in another roomThe warning exists, but only in one modality.
  • A platform announcement changes at a stationThe information is spoken once, then it is gone.
  • Someone knocks at the doorA quiet cue that is easy to miss from another room.
  • Instructions are given at speed in a clinicThe conversation moves faster than writing can follow.
02Four channels

One workspace.
Different ways to be understood.

Each module has one job and its own failure boundary. Start only what helps in the moment; the rest stays idle.

01Experimental

Sign to Voice

A camera-based module for a deliberately small, explicitly listed gesture set. It will never label an arbitrary pose as Indian Sign Language.

  • Real-time webcam tracking with local MediaPipe
  • Baseline ISL dataset recognition (Amrita SLR 49 classes)
  • Strict manual confirmation before speech synthesis
02Live

Voice to Text

Live captions for the speech around you, using the browser's speech recognition where it is available — with a manual typing fallback.

  • Live captions in English, Hindi, and Bengali where supported
  • Interim and final caption states
  • Clear notice when the browser uses an online service
03Experimental

Sound Alerts

On-device sound classification that surfaces possible alarms, doorbells, and other selected sounds — never a certified safety device.

  • YAMNet classification runs locally in the browser
  • Possible detections need repeated confirmation
  • Monitoring runs only while the page is open
04Live

Quick Communication Cards

Large, one-tap messages for everyday, medical, education, shopping, and emergency situations. This module works without camera or microphone.

  • English and Hindi speech; Bengali written text
  • No camera or microphone needed
  • Optional browser speech, never automatic
03How it works

Nothing starts without you.

SanketSetu is arranged as a workspace, not a feed. Every module is a tool you pick up and put down.

  1. 01

    Open the workspace

    No account, no setup wizard, no download screen. Open it in the browser and choose the channel that helps right now.

  2. 02

    Choose what helps now

    Quick Cards for a phrase you need immediately. Captions for speech. Sound awareness for the room. Gestures for a supported, explicitly listed set.

  3. 03

    Start, stop, and correct

    Permissions begin from a button. Status is always written out. Manual correction stays available, and a failing module never takes the workspace down.

One active channel at a time.

It keeps the browser honest about CPU, battery, and which device is listening.

04Privacy & access

Built for trust, not for collecting.

Accessibility and privacy are design constraints here, not marketing badges. The limits are part of the interface too.

Privacy

Local by default. Clear about the edges.

  • No account, no database, no analytics profile.
  • Camera frames and sound samples are designed to be processed in your browser, not uploaded by SanketSetu.
  • Captions are the exception: browser speech recognition may use an online service. We will say so before you start.
  • Model files download once and are cached by the browser.

Accessibility

Access is the interface, not a setting.

  • A keyboard path for every action, and a focus indicator that is never hidden.
  • Large touch targets, readable contrast, and status shown with words and icons.
  • Reading-size and reduced-motion controls are live in this build.
  • Hindi and Bengali content keeps its own language tag and script fallback.

Not a certified safety device

Future sound alerts and gesture recognition will be aids, never guarantees.

Speech depends on your browser

Caption language support and voice quality vary by browser and operating system.

AI modules are in development

This build ships the foundation, shell, and brand. The engines follow in later phases.

The first step

Ready when you are.

Open the workspace and see what works today. Quick Cards are live — no camera, no microphone, no model. The AI channels follow, one reviewed step at a time.