Real-time speech transcription and translation in your browser, powered by OpenAI's Realtime API.
Stream audio from any browser tab or microphone and get live subtitles with sub-second latency — with optional word-by-word streaming translation into 30+ languages, SRT export, and a floating subtitle overlay you can drag over any video player.
- Real-time transcription — word-by-word streaming via OpenAI Realtime API +
gpt-4o-transcribe - Streaming translation — translation tokens appear word-by-word as they are generated (LangGraph pipeline)
- Context-aware refinement — rolling topic summary and glossary improve transcription accuracy over time
- Semantic chat — ask questions about what's being said; answers are retrieved by semantic similarity across the entire session, not just the last few lines
- Floating subtitle overlay — detachable popup window you can position over any video on your desktop
- SRT export — download every session as a timestamped
.srtsubtitle file - Source language selection — lock to a specific language or let the model auto-detect
- VAD controls — tune speech sensitivity and silence duration via the advanced panel
- Tab or mic capture — capture any browser tab's audio or your microphone
- An OpenAI API key with access to the Realtime API
- Docker (recommended) or Python 3.12+
- A modern Chromium-based browser (for tab audio capture via
getDisplayMedia)
git clone https://github.com/stzifkas/whispersubs.git
cd whispersubs
cp .env.example .env
# edit .env and set OPENAI_API_KEY
docker compose up --buildOpen http://localhost:8000 in your browser.
To run in the background:
docker compose up -d --build
docker compose logs -f # follow logs
docker compose down # stopgit clone https://github.com/stzifkas/whispersubs.git
cd whispersubs
uv venv .venv && source .venv/bin/activate # or: python3 -m venv .venv
uv pip install -r requirements.txt # or: pip install -r requirements.txt
cp .env.example .env
# edit .env and set OPENAI_API_KEY
uvicorn backend.main:app --host 0.0.0.0 --port 8000 --reload- Select Source (Tab audio or Microphone) and optionally pick a Spoken language
- Pick a Translate to language if you want live translation
- Click Start — for tab audio, select a tab in Chrome's dialog and check Share tab audio
- Captions stream in real time. Click ⧉ for the floating subtitle overlay
- Click 💬 to open the live chat panel and ask questions about the content
- Click ↓ SRT to download the session transcript as a subtitle file
All settings have sensible defaults. Override via .env:
| Variable | Default | Description |
|---|---|---|
OPENAI_API_KEY |
— | Required |
REALTIME_MODEL |
gpt-4o-realtime-preview |
Realtime API transcription model |
WHISPER_MODEL |
gpt-4o-transcribe |
Transcription engine |
TRANSLATION_MODEL |
gpt-4o-mini |
Refinement + translation model |
CHAT_MODEL |
gpt-4o-mini |
Live chat model |
EMBEDDING_MODEL |
text-embedding-3-small |
Embeddings for semantic chat retrieval |
WHISPER_LANGUAGE |
(auto) | Lock source language (e.g. en, es) |
TARGET_LANGUAGE |
(none) | Default translation target |
NO_SPEECH_THRESHOLD |
0.6 |
VAD speech sensitivity (0–1) |
SILENCE_RMS_THRESHOLD |
200 |
Silence duration before turn ends (ms) |
Browser
└─ MediaStream (tab / mic) at 24 kHz
└─ ScriptProcessor → raw PCM16 binary frames
└─ WebSocket /ws (FastAPI)
├─ WebSocket → OpenAI Realtime API
│ └─ transcript deltas / completed events
├─ LangGraph pipeline (per completed turn)
│ ├─ refine_node — fix transcription errors (function_calling)
│ └─ translate_node — stream tokens word-by-word to browser
├─ Background: context extractor (rolling summary + glossary)
├─ Background: vector store (embeddings for semantic chat)
├─ In-memory SRT store (per session)
└─ Chat endpoint (semantic retrieval + rolling summary + GPT)
Contributions are welcome — bug reports, feature requests, and pull requests alike. See CONTRIBUTING.md for guidelines.
MIT © 2026 Sokratis Tzifkas