Voice settings
Set the language, voice, speech recognition, streaming, call limits, AI disclosure and recording policy for a Voice (IVR) project.
Per project, under Settings › Voice. These are the defaults every prompt inherits unless a node overrides them.
Voice output
Language and Voice for the whole flow. Individual prompts can differ, but most flows want one persona throughout. Leave Voice as the default to use the carrier's standard voice; the voices on offer come from the carrier your number belongs to.
Set the language even if you are speaking English: it decides pronunciation of numbers, dates and currency, and a number read in the wrong locale is noticeably wrong on a phone.
AI voice
Speaks prompts with a premium voice instead of the carrier's.
Pick an Engine, supply its API key, and choose a voice. The AI voice is applied when you publish. Prompts that contain variables keep the carrier voice for now.
Leave the engine on Carrier voice (default) to use the carrier's own text-to-speech.
Speech recognition
Transcribes call recordings with a speech-to-text vendor, including voicemails and calls on carriers with no speech recognition of their own. Choose an engine and supply its API key.
Transcribe every call transcribes each recording automatically as it lands. Off, you can transcribe individual calls from the call list in the project's Activity tab.
This engine is also what a flow listens with when streaming audio is on.
Streaming audio
Lets callers interrupt a prompt and be understood straight away, rather than waiting for it to finish. It needs both an AI voice and a speech-recognition engine, because streaming replaces the carrier's voice and recognition with your own, and a carrier that supports live audio.
See Streaming audio for setup.
Call limits
A runaway call bills by the minute until something stops it. Maximum call length (in seconds, default 600) ends the call at that point, whatever the flow is doing.
AI disclosure
Spoken once, before an AI Voice Agent first replies. Several jurisdictions require telling a caller they are speaking to a machine, and it is worth doing regardless. On by default, with editable wording.
Applies only to the AI Voice Agent node.
Recording calls
A conversation between an agent and a caller is recorded from the moment they are connected. Choose whether the person hears an announcement first (the default, with editable wording), Record without telling them, or Do not record agent calls. Someone an agent rings was never in your flow, so they have heard nothing unless this tells them.