Skip to content

About

Real-time multilingual code-switching transcriber web app with diarization, translation, playback, and chat.

Resources

Stars

1 star

Watchers

0 watching

Forks

Repository files navigation

Transcribee 🐝

Local-First, Real-Time Multilingual Code-switching Audio Transcriber, Diarization, Translation, Playback, & AI Chat.

Transcribee is a browser-only web application orchestrating real-time audio streaming, speaker diarization, live translation, audio scrubbing playback, LLM chat with FIFO context management, and offline persistence—completely client-side with zero intermediary backend servers.


📸 Interface Preview

View Description
Desktop Workspace
Desktop Workspace
Unified layout featuring saved session management, live transcript stream with speaker indicators, and dynamic AI chat drawer.
Transcription & Translation
Live Transcription
High-frequency token streaming with finalized text refinement, concurrent translation tracks, and click-to-play audio scrubbing.
Audio Playback & Scrubbing
Audio Playback
Dedicated playback session view with token scrubbing, speed controls (0.5x–2.0x), timeline seeking, and speaker synchronization.
System Configurations
System Configurations
Slide-out drawer for Soniox and Google Gemini API keys, model selections, audio routing modes, and language hints.
Mobile-Responsive Layout
Mobile Interface
Responsive viewport navigation collapsing auxiliary panels into fluid tabs focused on live transcription.

⚡ Core Features

  • Audio Capture & Loopback Mixing: Captures microphone (getUserMedia) and system audio (getDisplayMedia), mixing to 16kHz/44.1kHz linear PCM mono stream or routing single tracks.
  • Soniox Live STT Client: Full-duplex WebSocket client streaming real-time tokens, speaker diarization IDs, and translation tracks directly to Redux store.
  • Token-Level Audio Scrubbing: Correlates transcribed words to exact audio offsets (start_ms, duration_ms) for split-second audio playback from local cache.
  • Smart FIFO Context Manager: Client-side token counting with visual warning at 85% capacity and automated First-In, First-Out truncation at 100% preserving speaker turn boundaries.
  • In-Browser Gemini Chat: Context-aware assistant via LangChain JS streaming responses against accumulated session transcripts.
  • Local-First Persistence: Complete offline persistence for audio buffers, transcripts, translations, and chat histories via IndexedDB / OPFS.

🛠️ Architecture Overview

flowchart TD
    Audio["Web Audio API Processing<br/>(Mic + Speaker Loopback)"]
    RTK["RTK Global State Engine"]
    Soniox["Soniox Live STT Agent<br/>(WebSocket Engine)"]
    Gemini["Gemini Chat Agent<br/>(LangChain JS Client)"]
    Persistence["Local-First Persistence Agent<br/>(IndexedDB / OPFS)"]

    Audio -->|"Raw Audio Buffer (PCM)"| RTK
    RTK -->|"Audio Stream"| Soniox
    Soniox -->|"Real-Time Tokens & Diarization"| RTK
    RTK -->|"Context Stream"| Gemini
    Gemini -->|"Streamed Tokens"| RTK
    RTK -->|"Structured Payload & Audio Blobs"| Persistence
Loading

🚀 Quickstart

Prerequisites

  • Node.js (v18+)
  • npm or bun

1. Install Dependencies

npm install
# or
bun install

2. Start Development Server

npm run dev
# or
bun run dev

Open http://localhost:3000.

3. Running End-to-End Tests

Playwright validates full user workflows (recording, transcription, translation, playback, AI chat, persistence):

npm test
# or
bun test

About

Real-time multilingual code-switching transcriber web app with diarization, translation, playback, and chat.

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages