request-session

Example: Request Voice Session

Request a real-time speech translation session token from the DeepL Voice API.

Files

File	Description
curl.sh	Minimal curl reference command (session request only)
node/	Node.js implementation
python/	Python implementation

Quick start

# Node.js
cd node && DEEPL_API_KEY=your-key node index.js

# Python
cd python && DEEPL_API_KEY=your-key python main.py

What it does

1. Reads DEEPL_API_KEY from the environment.
2. Calls POST /v3/voice/realtime with the desired audio format (audio/pcm;encoding=s16le;rate=16000), source language, target languages, and message format.
3. Receives a streaming_url, ephemeral token, and session_id.
4. Prints the WebSocket connection URL (streaming_url?token=…) to use for streaming.

Prerequisites

• A DeepL API Pro subscription with Voice API access (Enterprise only — contact sales)
• Node.js >= 18 or Python >= 3.8

Sample response

{
  "streaming_url": "wss://api.deepl.com/v3/voice/realtime/connect",
  "token": "VGhpcyBpcyBhIGZha2UgdG9rZW4K",
  "session_id": "4f911080-cfe2-41d4-8269-0e6ec15a0354"
}

Connect to the WebSocket using:

wss://api.deepl.com/v3/voice/realtime/connect?token=VGhpcyBpcyBhIGZha2UgdG9rZW4K

Notes

• The session token is ephemeral and single-use; request a new session for each stream.
• The WebSocket streaming protocol is documented at developers.deepl.com/api-reference/voice/websocket-streaming.
• Supported audio formats include PCM, FLAC, MP3, OPUS, and AAC; see the API reference for the full list.
• A maximum of 5 target languages can be specified per session.
• Streams time out after 30 seconds of silence and have a maximum duration of 1 hour.
• Audio chunks must not exceed 100 KB or 1 second of audio; 50–250 ms chunks are recommended.