Overview
BhashaSethu (“language bridge”) is a mobile-first progressive web app for live Hindi-to-Kannada voice translation. It’s built for travellers who speak Hindi and need to communicate in Karnataka, using Sarvam AI’s Indic language models for speech-to-text, translation, and text-to-speech in a single press-and-hold interaction.
Tech stack
- Framework: Next.js 16 with App Router
- Language: TypeScript
- UI: React 19, Tailwind CSS v4, shadcn/ui
- State management: Zustand
- AI: Sarvam AI SDK — speech-to-text (Saaras v3), translation (Mayura v1), text-to-speech
- PWA: installable, offline-ready
How it works
- Press and hold the mic button to record speech in Hindi or Kannada
- Sarvam AI transcribes the audio with the Saaras speech-to-text model
- The Mayura translation model converts the text to the target language
- Text-to-speech plays the translation aloud for the listener
The app caches TTS audio, so common phrases don’t cost repeated API calls.
Features
- Press-and-hold recording with automatic transcription, translation, and playback
- Works both directions: Hindi to Kannada and Kannada to Hindi
- Phrasebook of common phrases with tap-to-speak
- One-tap quick cards for everyday situations
- History of recent translations
- Cached TTS audio for speed and fewer API calls
- Automatic dark and light mode
- Installable PWA designed for one-handed thumb use
- Sarvam AI supports 11 Indian languages: Bengali, English, Gujarati, Hindi, Kannada, Malayalam, Marathi, Odia, Punjabi, Tamil, and Telugu
Why I built this
Moving to Bangalore as a Hindi speaker, everyday interactions with auto drivers, shopkeepers, and neighbours often hit a language wall. Existing translation apps feel clunky for quick spoken exchanges. BhashaSethu is built for exactly this: hold the button, speak, and let the other person hear the translation.
Source code on GitHub. Try it at local-guide.abhimanyusikarwar.com.