# Speech Input IVR: Capture Spoken Answers on a Call

> How a speech input IVR captures what callers say: the Ask node, timeouts, re-asking, expected words, Hindi and English, where answers go, and when to use keypad input instead.

Source: https://edesy.in/blog/speech-input-ivr-capture-spoken-answers-2026

**A speech input IVR asks a question and records the caller's spoken answer as text. In Edesy IVR Studio
this is the Ask node: it speaks the question, waits for an answer, transcribes it in English or Hindi,
re-asks if the caller is silent, and stores the answer under a key you name for later steps and reports.**

> **Add a spoken question to your flow:** [Try Edesy IVR Studio](https://ivr.edesy.in/?utm_source=edesy.in&utm_medium=blog&utm_campaign=speech-input-ivr-capture-spoken-answers-2026) — Rs 20 free credit on sign-up, no card needed.

## Keypad vs speech: when to use which

| Question type | Better input | Example |
|---|---|---|
| Fixed choice, 2 to 4 options | Keypad | "To confirm, press 1. To cancel, press 2." |
| Rating | Keypad | "Rate us from 1 to 5." |
| Number (PIN, order ID) | Keypad | "Enter your 6-digit order number." |
| Place or name | Speech | "Which city are you calling from?" |
| Open reason | Speech | "In a few words, why are you calling today?" |
| Free-form feedback | Speech | "What could we have done better?" |

Keypad input is exact and works on any phone. Speech handles answers that won't fit on 12 keys. For
keypress basics, see [what is DTMF](https://edesy.in/blog/what-is-dtmf-ivr-keypress-guide-2026).

## How the Ask node works

1. **Speak the question** in the chosen language (text-to-speech, with `{{var.name}}` tokens allowed).
2. **Listen** for up to the wait time.
3. **End the answer** when the caller pauses for the end-of-speech time.
4. **Transcribe** the answer with speech-to-text.
5. **Re-ask** (optionally with different wording) if the caller said nothing.
6. **Exit** on **answered** or **no answer**.

## Ask node settings

| Setting | Range / default | Tip |
|---|---|---|
| Question key | lowercase letters, digits, `_` | Names the answer: `city` becomes `{{answer.city}}` |
| Question | text, with audio preview | One question, plainly worded |
| Language | English (India), English (US), Hindi | Used both to speak and to transcribe |
| Wait for an answer | 1 to 30 s, default 8 | Longer for open questions |
| Pause that ends the answer | 1 to 5 s, default 2 | Raise it for reasons and feedback; lower it for one-word answers |
| Re-ask if silent | 0 to 2, default 1 | Extra attempts before taking the no-answer exit |
| Re-ask prompt | optional text | Shorter and simpler than the original question |
| Expected words | comma-separated | Names, places and terms callers are likely to say |

## Writing questions that transcribe well

- **Ask one thing.** "Which city?" not "Which city and which branch?"
- **Hint the shape of the answer.** "In one or two words, what is this about?"
- **Use expected words** for anything with spelling variants: branch names, product names, cities.
- **Plan for silence.** A re-ask prompt like "Sorry, please tell me the city name" gets a second chance.
- **Don't ask for numbers by voice** when the keypad will do. Digits are more reliable typed than spoken.
- **Hindi:** write the question in Devanagari or natural Hinglish, as callers speak it. Example placeholder
  from the builder: "Aap kahaan se baat kar rahe hain?"

## Where answers go

- **Call log:** each answer appears on the IVR call log with the call's variables and recording.
- **Status callback:** for calls placed through the outbound IVR API, answers are included in the signed
  callback when the call ends. See the [outbound IVR API guide](https://edesy.in/blog/ivr-outbound-call-api-guide-2026).
- **Later nodes:** `{{answer.city}}` works in an HTTP Request body or headers and in an Email notification.

## Acting on what was said

The Ask node captures the answer; it doesn't interpret it. Three patterns:

1. **Store and review.** Survey reasons, callback requests, feedback. Read them in call logs or export them
   from your own system via the callback.
2. **Send to your API live.** An **HTTP Request** node posts `{{answer.reason}}` to your endpoint. In sync
   mode the caller waits up to 5 seconds and the flow branches on Success or Failure, so your API can
   decide, for example, whether the city is one you serve. See the
   [HTTP Request node guide](https://edesy.in/blog/ivr-http-request-node-call-your-api-2026).
3. **Notify a person.** An **Email notification** node sends the answer and caller number to the workspace
   owner.

If you need a full back-and-forth conversation rather than single questions, that's an AI voice agent, not
an IVR. IVR Studio can hand the call to your own bot with the Stream node, or see the
[AI voice agent](https://edesy.in/ai-voice-agent).

## Example: post-service survey with a reason

1. **Say:** "Thanks for visiting {Business} today."
2. **Menu:** "Please rate us from 1 to 5." (keypad)
3. **Ask** (key `reason`): "In a few words, what's the main reason for your rating?" End-of-speech 3 s,
   re-ask once.
4. **HTTP Request** (async): POST `{"rating": "{{digits}}", "reason": "{{answer.reason}}"}` to your endpoint.
5. **Say:** "Thank you. Goodbye." **Hangup**.

The [keypad survey guide](https://edesy.in/blog/keypad-ivr-survey-nps-guide-2026) covers the rating half in more detail.

## Testing

Use the audio preview to hear each question without a call, then place a test call and answer in the way
real callers will: quickly, slowly, with background noise, and in Hinglish if that's how your customers
talk. Check the transcript in the call log and add expected words where it misheard.

## Cost

Ask nodes are part of the call: IVR Studio is Rs 1.50/min, plus number rental, excluding 18% GST. See
[IVR pricing in India](https://edesy.in/blog/ivr-pricing-india-2026). Feature list on the [speech input feature page](https://edesy.in/ivr/features/speech-input),
setup in the [IVR docs](https://edesy.in/docs/ivr).

[Add an Ask node to your flow](https://ivr.edesy.in/?utm_source=edesy.in&utm_medium=blog&utm_campaign=speech-input-ivr-capture-spoken-answers-2026)
with Rs 20 free credit, no card needed.

## Frequently Asked Questions

### What is a speech input IVR?

An IVR that asks a question and listens to the caller's spoken answer instead of, or as well as, waiting for a keypress. The answer is transcribed to text by speech-to-text and saved with the call.

### Which languages can the Ask node transcribe?

The Ask node's language list currently offers English (India), English (US) and Hindi. The same language is used to speak the question and to transcribe the answer.

### Can the IVR branch on what the caller said?

The Ask node itself has two exits: answered and no answer. To act on the content, send {{answer.<key>}} to your own API with a sync HTTP Request node and branch on its Success or Failure result, or process the answers afterwards from the call log or status callback.

### How do I improve recognition of names and places?

List them in the Expected words field, comma-separated, for example 'Delhi, Lucknow, Sales, Accounts'. The hint tells the recogniser what callers are likely to say.

### Where do the spoken answers end up?

In the IVR call log next to the recording, in the status callback for outbound API calls, and in later nodes as {{answer.<question key>}} tokens, for example in an HTTP Request body or an email notification.

### Is speech input better than keypad input?

Not always. Use keypad input for fixed choices (confirm or cancel, a 1 to 5 rating) and speech for open answers that don't fit on a keypad, such as a city, a reason, or a name. Mixed flows often work best.