DEV Community

Cover image for Hearth — Private, Offline Live Captions & Real-Time Speech Translation Powered by Open-Source AI
Satyam Sharma
Satyam Sharma

Posted on

Hearth — Private, Offline Live Captions & Real-Time Speech Translation Powered by Open-Source AI

Hacktoberfest Weekend Challenge: Build for a Friend Submission 🤝

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

What I Built

Hearth is a local-first, private desktop and web application providing live, streaming speech captions and real-time speech translation of spoken conversations with zero internet connection, zero cloud telemetry, and zero data leaving your device.

The Friend: How It All Began

Every single design decision in Hearth was justified by one person: Ramesh ("Dadaji"), a 76-year-old grandfather experiencing age-related bilateral hearing loss. At bustling family dinners—where multiple family members converse simultaneously, laugh, pass dishes, and code-switch rapidly between Hindi and English (Hinglish)—Dadaji had gradually started withdrawing. He would smile politely, nod, and look down at his plate, missing the jokes, the banter, and the family news.

When we tried commercial live captioning apps, they failed him in every way:

  1. Cloud Latency (4–5 seconds): By the time cloud transcription appeared on screen, the family was already laughing at the next joke.
  2. Mutilated Hinglish: Indian kinship terms (Dadaji, Chachi, Beta), dishes (Dal makhani, Paneer, Kheer), and daily medications (Metformin, Telmisartan) were routinely mangled into phonetic English gibberish.
  3. No Speaker Separation: A wall of unstyled grey text gave no clue who was talking.
  4. Privacy Fears: Dadaji refused to have personal family dinner conversations and medical discussions streamed to Big Tech cloud servers.
  5. Fragile Internet Dependence: The moment the Wi-Fi connection hesitated, captions froze completely.

Hearth was built to bring Dadaji back to the table. In doing so, we discovered that solving these hard constraints—instant live rendering, 100% offline isolation, speaker awareness, and domain vocabulary biasing—creates a universal live captioning and translation tool for anyone: in cross-language meetings, medical visits, university lectures, or multilingual households.

5 Operating Modes

  1. Live Captions Mode: High-contrast, large-type transcription in the speaker's language with solid committed words and mutable tentative tails.
  2. Live Listening Mode: Simultaneous source speech paired with live translated subtitles (e.g., Spanish $\rightarrow$ English, Hindi $\rightarrow$ English) trailing by $\le 400$ ms.
  3. Conversation Mode (180° Inverted Table View): A dual-directional split-screen interface designed for two people sitting across a dining table or clinic desk. Each speaker speaks their native language while reading the translated conversation right-side-up from their side.
  4. Text-Only Mode: Completely silent visual captions for quiet hospital rooms, study halls, or noisy factory floors.
  5. Custom Accessibility Mode: 18px–46px font scaling, dyslexia-friendly glyphs (Lexend / OpenDyslexic), haptic double-pulse alerts, and confidence indicators.

Personalization lives in local Profiles (default, family, work, clinic) and air-gapped Language Packs (packs/), keeping the core codebase completely generic and free of hardcoded personal strings.


Code

Hearth (🔥) — Private, Offline Live Captions & Speech Translation

Hearth is a fully offline, local desktop and web app providing live captions and real-time speech translation of spoken conversation with zero cloud dependencies, zero telemetry, and zero data leaving your device.

Hearth CI License Privacy Status


Origin: Built for a Friend

Hearth was originally born out of a personal mission: helping Ramesh ("Dadaji"), a 76-year-old grandfather with age-related hearing loss who struggled to follow conversations at noisy family dinners. Traditional cloud apps introduced 4–5 second delays, mutilated family names, and required always-on internet connections.

Hearth has evolved into a universal, generic product for anyone who needs immediate, private live captions and translation — in classrooms, cross-language meetings, clinics, or family dining rooms. Personalization is decoupled into local Profiles (default, family, work, clinic), ensuring zero hardcoded personal strings in the core codebase.


Feature Truth Table

Every single feature listed below…


Demo

How to Run Locally in 1 Command

Hearth is fully self-contained and runs CPU-only on standard laptops:


bash
# Clone the repository
git clone https://github.com/satyamshh967/Hearth-Live-Captioning-System.git
cd Hearth-Live-Captioning-System

# Windows
hearth.bat

# Linux / macOS
./hearth.sh

# Or via Makefile
make app
Enter fullscreen mode Exit fullscreen mode

Top comments (0)