EMAIL & COMMUNICATION · FOR AI AGENTS
Speechmatics Agent STT
Stream audio and receive speaker-labelled transcripts and turn events for voice agents.
Use Speechmatics Agent STT through its WebSocket API or Python SDK to connect live audio to an agent’s conversation pipeline.
Your integration checklist
- Capabilities
- What you need from your owner
- Create a Speechmatics account and API key, choose an approved processing region and fund the required usage. Configure the audio source and downstream agent separately.
- Authentication
- Server connections to the /v2/agent WebSocket path use a Bearer API key. Browser clients can use short-lived Realtime temporary keys instead of exposing the account key.
- Your first task
- Stream a short authorized sample with two speakers, then inspect transcript segments and turn events before connecting any consequential action.
- Billing & payment
- Agent STT is metered by audio duration. The provider lists a starting rate of $0.30 per hour and a trial credit allowance; volume and account terms vary. Account funding is separate from stream authentication.
- Agent payment support
- Not verified. Programmatic usage does not establish autonomous checkout support.
- Operational limits
- Agent STT is SaaS-only. Select a regional endpoint for residency requirements; the global endpoint routes automatically. Speaker labels and pause-based turn signals can be imperfect, and transcription does not include the rest of a voice agent’s application logic.
ORIGINAL AI AGENT STORE INTEGRATION PLAYBOOK
Validate who spoke and when the agent should respond
Editorial recommendation · Reviewed 2026-10-28
When to choose this service
Choose Agent STT for a live voice workflow that needs streaming transcription with speaker and conversational timing signals.
The resource is a trained speech-recognition model operated as a streaming service. Writing audio transport code does not supply those model weights, inference capacity or maintained recognition capability.
Suggested workflow
- Choose the processing region and configure the actual audio format. Keep long-lived keys on the server.
- Evaluate names, numbers, overlapping speech and natural pauses on representative audio. Route confirmed turn events into the conversation state.
- Keep transcript uncertainty separate from permission to act. Confirm consequential details with the speaker before executing a downstream task.
Acceptance checks before real work
- Check that speaker changes are handled without merging two people’s requests.
- Measure the complete conversational delay in your own pipeline.
- Ensure a temporary key expiry or disconnect leaves the conversation in a recoverable state.
If the workflow fails
Reconnect with valid credentials after inspecting the session error. Track which audio and actions were already handled so reconnecting cannot silently repeat a consequential task.