Prefer code? Configure Voice through the API. Shared behavior is explained in Voice configuration.
Before you begin
- A voice agent and project edit access.
- A test number assigned to your draft, or the custom audio client you will deploy.
- Representative requests, including a pause, interruption and any keypad entry.
Configure and test Voice
Configure speech recognition
Open the agent’s Behaviour → Voice input. Choose the recognizer and input language. Adjust its available end-of-turn and interruption controls only after reproducing the timing problem on a test call.
Configure spoken output
In Voice output, choose the default voice. Add Language overrides for the locales you use. Open Pronunciation dictionary for terms the selected voice mispronounces. Configure Audio cue, Background audio or Sync audio and text output when your call needs them.
Configure call controls
For phone projects, enable Keypad input and choose completion keys and an inter-digit timeout. Ask for the digits in your agent or input step. Where available, leave Listen for other triggers on. Test that the submitted value reaches the expected variable; the runtime submits at 20 digits.
In Session & timeout, set Inactivity timeout, Max conversation duration and an optional End message. Choose Remember callers across calls according to whether variables should persist.
Verify the saved behavior
Reopen the settings to confirm their saved values. Call the draft and test one change at a time: pronunciation, natural pauses, interruption, keypad submission, silence and duration. Call twice to check the intended returning-caller behavior.
Verify the result
The recognizer captures the request; the selected voice pronounces your test terms correctly; interruptions clear old audio; collected digits reach their intended variable. Silence and duration checks end the call under their respective conditions.
Revert a configuration change
Restore the previous setting if the change worsens the reproduced issue. Publish only after the actual connection passes the checks in Release a voice agent.