Give your companion a voice in 10 minutes

The fastest way to make a companion feel real is to give it a voice. Text is a conversation; voice is a presence. Here's the 10-minute version using the SecretVoice API.

1. Get a key

Sign up, grab your API key from the dashboard, and you're ready. Every account ships with a default key right away, and verified teams can request a free trial grant.

2. The one-liner (synchronous)

curl -X POST https://api.secretvoice.dev/v1/text-to-speech/velvet \
  -H "xi-api-key: $SECRETVOICE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"text": "There you are. I was starting to miss you.", "model_id": "secret-v1"}' \
  --output hello.mp3

3. Realtime (WebSocket) for live turns

When your companion should respond, not just narrate. Use the realtime stream:

wss://ws.secretvoice.dev/v1/text-to-speech/velvet/stream-input

Send text chunks as they arrive from the LLM, flush, and audio frames stream back with sub-200ms time-to-first-audio. Pipe the frames into a WebRTC or LiveKit session and the companion is talking live.

4. Keep the persona locked

The real magic for companions is persona stability: zero acoustic drift over multi-turn sessions. The voice your user fell for in message two is the same voice at message two thousand: same timbre, same intimacy, same pacing. That's what makes a companion feel like one person, not a generator.

Switch voices at runtime (the API accepts any of the 29), but the character stays yours.

What builders ship with this

Voice notes in companion apps (send her a message, she whispers a reply)
Realtime companion calls over WebSocket
Telegram bots that answer in voice
Roleplay sessions where the character never breaks voice

That's it: key, one endpoint, one WebSocket. Go build the companion, and let her speak for herself.