PolyDub: Real-Time Multilingual Video Dubbing

Speak any language. Be heard in any language. Live.

PolyDub is a browser-based multilingual dubbing platform for live broadcasts, video rooms, and uploaded media. It captures speech, transcribes it with Deepgram, translates it, generates dubbed speech, and plays it back to listeners in their selected language.

Features

Mode	Route	What it does
Live Broadcast	/broadcast, /broadcast/[id]/[lang]	One host speaks while listeners receive translated transcripts and dubbed audio.
Multilingual Rooms	/rooms, /room/[roomId]	Multiple participants speak in their own language and hear others in their selected language.
VOD Dubbing	/vod	Upload media and generate dubbed audio, SRT subtitles, and a muxed MP4.

Highlights:

• Low-latency browser dubbing target for live speech.
• Native Aura-2 voice options for English, Spanish, French, German, Italian, Japanese, and Dutch.
• Per-listener TTS queues so generated room audio does not interleave across speakers.
• English + 11 translated UI locales compiled at build time with Lingo.dev.
• VOD pipeline for prerecorded transcription, translation, TTS audio, SRT generation, and FFmpeg muxing.

Screenshots

Landing Page

Broadcast Mode

Multilingual Room

VOD Dubbing Studio

How It Works

See ARCHITECTURE.md for sequence diagrams, WebSocket role routing, and pipeline details.

Mic PCM -> WebSocket
-> Deepgram Nova-2 STT
-> Google Translate gtx adapter
-> Deepgram Aura-2 TTS
-> WebSocket -> Browser AudioContext -> Speaker

PolyDub runs as two cooperating processes:

Next.js app (:3000)
- /broadcast
- /broadcast/[id]/[lang]
- /rooms
- /room/[roomId]
- /vod
- /api/dub
- /api/mux
- /api/tts-preview
WebSocket server (:8080)
- broadcast hosts and listeners
- room audio and video sockets
- Deepgram streaming STT
- translation cache and in-flight deduplication
- Deepgram streaming TTS

Supported Languages

Live/VOD source language options: English, Spanish, French, German, Italian, Dutch, Japanese, Portuguese, Hindi, Arabic, Korean, Turkish, Vietnamese, Ukrainian, and Polish.

Aura-2 target voices exposed in the UI:

Language	Voices
English	Thalia, Andromeda, Apollo, Arcas
Spanish	Celeste, Estrella, Nestor, Sirio
French	Agathe, Hector
German	Viktoria, Elara, Julius, Fabian
Italian	Livia, Melia, Dionisio, Elio
Japanese	Izanami, Uzume, Ama, Ebisu, Fujin
Dutch	Rhea, Beatrix, Sander, Lars

Tech Stack

• App: Next.js 16, React 19, TypeScript
• UI: Tailwind CSS v4, shadcn-style UI primitives, Radix UI, Phosphor Icons
• Realtime: Node.js, ws, browser Web Audio APIs
• STT: Deepgram Nova-2 streaming and prerecorded transcription
• Translation: Google Translate unofficial gtx endpoint with LRU cache and in-flight deduplication
• TTS: Deepgram Aura-2
• VOD muxing: System ffmpeg binary invoked by /api/mux
• UI i18n: Lingo.dev compiler
• Testing: TestSprite-generated backend and frontend tests

Prerequisites

• Node.js 18 or newer
• pnpm
• A Deepgram API key for STT and TTS
• A Lingo.dev API key for UI translation compilation
• FFmpeg installed and available as ffmpeg on PATH for production VOD muxing

Getting Started

git clone https://github.com/crypticsaiyan/polydub.git
cd polydub
pnpm install
cp .env.example .env

Edit .env:

DEEPGRAM_API_KEY=your_deepgram_api_key_hereLINGO_API_KEY=your_lingo_api_key_hereLINGO_BUILD_MODE=translatePORT=8080WEBSOCKET_PORT=8080NEXT_PUBLIC_WS_URL=ws://localhost:8080

Start both processes in separate terminals:

pnpm dev

pnpm run server

Open http://localhost:3000.

Scripts

Command	Description
pnpm dev	Start the Next.js dev server on http://localhost:3000.
pnpm run server	Start the TypeScript WebSocket server on port 8080.
pnpm build	Build the Next.js app and run Lingo compilation.
pnpm start	Start the built Next.js app.
pnpm server:build	Compile the WebSocket server TypeScript to dist-server/.
pnpm server:start	Start the compiled WebSocket server from dist-server/index.js.
pnpm lint	Run ESLint.

Environment Variables

Variable	Used by	Required	Description
DEEPGRAM_API_KEY	Next.js API routes, WebSocket server	Yes	Deepgram STT and TTS API key.
LINGO_API_KEY	Next.js build, /api/dub guard	Yes for full build/VOD flow	Mapped to LINGODOTDEV_API_KEY in next.config.mjs for Lingo.dev compilation.
LINGO_BUILD_MODE	Next.js build	No	Optional Lingo compiler mode. Defaults to translate.
PORT	WebSocket server	No	Primary WebSocket server port. Defaults to 8080.
WEBSOCKET_PORT	WebSocket server	No	Fallback port when PORT is not set.
NEXT_PUBLIC_WS_URL	Browser client	Yes	Public WebSocket URL, for example ws://localhost:8080 or wss://your-ws-domain.com.

API Reference

POST /api/dub

Accepts multipart form data and returns generated subtitles plus base64 audio.

Field	Required	Description
file	Yes	Uploaded audio/video file.
targetLanguage or target_language	Yes	Target language code for dubbing.
sourceLanguage or source_language	No	Source language code. Defaults to auto.
voiceId or voice_id	No	Deepgram Aura-2 voice ID override.

Response:

{
"srt": "1\n00:00:00,000 --> 00:00:01,500\n...",
"mp3": "base64-encoded-audio"
}

POST /api/mux

Accepts multipart form data and returns a video/mp4 response.

Field	Required	Description
video or original_video	Yes	Original video file.
audio or dubbed_audio	Yes	Dubbed audio file.

POST /api/tts-preview

Returns preview audio for a short text/voice sample used by the voice selector UI.

Project Structure

app/
broadcast/ Live broadcast host page
broadcast/[id]/[lang]/ Live broadcast listener page
room/[roomId]/ Multilingual room page
rooms/ Room lobby
vod/ VOD dubbing studio
api/dub/ STT -> translate -> TTS -> SRT/audio
api/mux/ FFmpeg audio/video muxing
api/tts-preview/ Voice preview endpoint
components/polydub/ Core product UI components
components/room/ Room video components
components/ui/ Shared UI primitives
hooks/ WebSocket and responsive UI hooks
lib/srt.ts SRT subtitle generation
server/ WebSocket server and Deepgram/translation wrappers
testsprite_tests/ Generated TestSprite tests and reports

Testing

Tests are generated and executed using TestSprite MCP.

Initial Round: Issues Found

TestSprite surfaced two concrete bugs on the first run:

1. TC003 (POST /api/dub third-party failure): The error handler was returning a plain text string on API failures instead of a JSON response. This caused clients to receive an unparseable body on 500 errors.
2. TC009 (malformed room ID): The frontend form was submitting malformed room IDs to the server without client-side validation. The server rejected them but no user-visible error was shown.

Both were fixed before Round 2.

Final Rounds: All Tests Passing

Backend API Tests (5/5 Passing)

Test	What it checks
TC001	POST /api/dub returns JSON with non-empty srt and valid base64 mp3.
TC002	POST /api/dub missing params returns 400 with Missing parameters.
TC003	POST /api/dub third-party failure returns 500 with an error body.
TC004	POST /api/mux valid video + audio produces a video/mp4 stream.
TC005	POST /api/mux missing inputs returns 400 with Missing video or audio file.

Frontend / E2E Tests (12 Cases Passing)

Test	What it checks
TC001	Start live broadcast after selecting source and target languages.
TC002	Create a new room and reach the in-room view.
TC003	Join an existing room by ID.
TC004	Leave and rejoin a room with the same ID.
TC005	Prevent broadcast start when no target languages are selected.
TC006	Change language and voice selection in an active room.
TC007	VOD studio opens in upload-ready state.
TC008	Navigate from landing page to broadcast setup.
TC009	Show validation when joining with a malformed room ID.
TC010	Navigate from landing page to rooms lobby.
TC011	Reject unsupported VOD file type on upload.
TC012	Navigate from landing page to VOD studio.

Test files are in testsprite_tests/.

License

MIT