Skip to content

How to add voice verification to a Twilio voice agent

Fork the caller’s audio with Twilio Media Streams, score it in real time with the Vosco API, and gate sensitive actions on the trust result. A step-by-step pattern.

Vosco Research3 min read
IntegrationsDevelopersVoice agents

The pattern

Most production voice agents on Twilio already receive call audio for speech-to-text. Voice verification uses the same idea: send a copy of the caller’s audio to a scoring service in parallel, and read the result back before the agent does anything sensitive. Nothing about the conversation flow has to change.

Step 1 — fork the call audio

Twilio’s <Stream> instruction inside <Start> streams the call’s audio to a WebSocket URL asynchronously, so the call keeps running while audio flows to the stream.

<Response>
  <Start>
    <Stream url="wss://api.vosco.ai/api/v1/connect/stream/twilio?install=INSTALL_REFERENCE"
            track="inbound_track" />
  </Start>
  <!-- …connect the call to your voice agent -->
</Response>

INSTALL_REFERENCE comes from the Integrations screen of the Vosco dashboard. Your API key is stored, encrypted, on that install — so no credentials go into TwiML and there is no relay service to run.

Step 2 — read the trust signals

As audio arrives, Vosco scores the caller and posts a signed verdict to your webhook: live, replay, synthetic or undetermined. Checks keep running for as long as the call lasts — so it keeps confirming the same caller, and a mid-call swap to a recording or a clone is caught too.

Step 3 — gate the action, not the call

Keep ordinary turns frictionless. Before a transfer, a payment or a credential reset, check the latest result:

  • Allow — the verdict is live.
  • Challenge — ask a follow-up, send a one-time code, or move to a stronger factor.
  • Deny — synthetic or replayed audio; block and log.
  • Escalate — hand to a human with the per-signal breakdown attached.

Step 4 — enroll and test

Speaker verification needs an enrolled voiceprint for each customer; liveness, replay and synthetic detection work without enrollment. Test with your own recordings — live calls, replayed audio and generated speech — at your telephony codec and operating point before going live. The developer page has the quick start, and you can start with 1,000 minutes free every month.

Other platforms

The same approach works beyond Twilio. See caller verification in Vapi, LiveKit and Pipecat, or call verification as a tool through the Vosco MCP server.

Frequently asked questions

How do I stream Twilio call audio to another service?

Use Twilio Media Streams: a Stream instruction inside Start in your TwiML sends the call’s audio to a WebSocket URL while the call continues.

Does adding voice verification change my Twilio call flow?

No. The audio is forked in parallel; your agent keeps handling the conversation and only checks the trust result before sensitive actions.

How much does it cost?

The Vosco API is usage-based at $0.0042 per minute, lower at volume, with 1,000 minutes free every month.

Add caller verification to your Twilio agent.

Fork the call to the Vosco API and gate sensitive actions on a live trust score. 1,000 minutes free every month.