IVR node: Voice recognition

The Voice recognition node plays a prompt and records a spoken reply from the caller during a short window. The audio is automatically transcribed and the text becomes a variable you can use later in the flow (for example in a Branch or Condition node). It's useful when asking the caller to say something ("Tell us your name", "How can we help you?") is more natural than asking them to press a number.

Requirements

• The workspace must have the voice transcription service this node relies on enabled.
• Decide in advance which variable will hold the recognized text so you can reference it in later nodes.

Steps

1. Open Settings → Channels → (your calling channel) → IVR workflow editor.
2. Drag the Voice recognition node onto the canvas.
3. Write the Prompt that plays before recording (e.g. "Briefly tell us how we can help you").
4. Pick a voice and set the Max duration.
5. Fill in Save transcription to variable with a name (e.g. call_reason).
6. Optional: set a Language hint if you already know the caller's language.
7. Leave Fall back to DTMF if recognition fails on unless you have a different plan for that case.
8. Connect the output to the next node (e.g. a Branch using the saved variable, or straight to Route to AI).
9. Save and publish.

Configuration

• Prompt: what's asked of the caller before recording.
• Max duration (seconds): how long it records after the prompt. Hard cap of 30 seconds, because the transcription step has a 10-second wait limit.
• Save transcription to variable: the name of the variable that holds the recognized text, for use later in Branch, Condition, or Route to AI.
• Language hint (ISO 639-1): a code like "es", "en", or "pt"; leave empty for auto-detection.
• Fall back to DTMF if recognition fails: if transcription comes back empty or fails, it's a good idea to route the call to a digit-based prompt as a fallback.

Example

In your opening greeting you ask "How can we help you?" with Voice recognition, save the answer to call_reason, and a Route to AI node uses that variable as context so the AI agent understands the request without asking the caller to repeat everything.

Tips

• Pair it with a Route to AI or Branch node that reads the saved variable to make decisions based on what the caller said.
• If you expect short, predictable answers (numbers, menu options), Wait for digit (DTMF) is usually more reliable than this node.

Troubleshooting

• Transcription comes back empty: the caller didn't speak, spoke too quietly, or the audio didn't arrive in time within the 10-second limit — make sure the DTMF fallback path is wired up.
• The variable shows up empty in later nodes: confirm the name in Save transcription to variable matches exactly what you use afterward.