IVR Prompts: Recording a Voice vs Text-to-Speech Compared
How to choose between a recorded human voice and text-to-speech for phone prompts, how to write for the ear, and how to keep audio clear and current.
Every phone menu is mostly words said aloud. How those words sound shapes the caller's first impression, and how quickly you can change them decides whether they stay accurate. The central choice is between a recorded human voice and a synthetic one, usually called text-to-speech, or TTS.

This guide compares the two, explains how to write prompts that are easy to follow, covers recording and synthetic-voice practice, and ends with a maintenance routine. It supports IVR Menu Design and 10 Common IVR Mistakes and How to Fix Them.
Recorded voice or text-to-speech
Comparing the two approaches
| Recorded human voice | Text-to-speech | |
|---|---|---|
| Sound | Warm, natural, with the speaker's personality | Consistent and clear; quality varies by voice and language |
| Changing a prompt | Needs a new recording session, or the same speaker and room | Edit the text and generate again |
| Names and unusual words | Read correctly by someone who knows them | May need pronunciation adjustments |
| Languages | You need a speaker for each | Often easy to add, but check quality for each language |
| Cost and effort | Time to record, edit and re-record | Time to write and check; little production work |
| Risk | Voice changes if the speaker leaves | Mispronunciations, flat delivery in long passages |
Neither is superior in every case. A brand-new restaurant may like a friendly recording of the owner saying "Thank you for calling." A property manager with seasonal notices that change weekly will value being able to edit text.
The hybrid approach
A common compromise is to record the greeting and closing, which rarely change, and to use TTS for hours, notices, names and anything that updates. Keep the voices similar in tone so the join is not jarring, and test the transitions.
Writing for the ear
Text that reads well on a page can sound clumsy when spoken. Read every prompt aloud before you record or generate it.
- One idea per sentence. Short sentences are easier to follow.
- Put the label before the key. "For billing, press 1."
- Write numbers the way they are spoken. Spell out dates and times in a way the voice will read correctly, and check how the system reads them.
- Avoid abbreviations. Write out the full word unless everyone says the abbreviation aloud.
- Add pauses on purpose. A beat after each option gives callers time to think.
- Cut filler. "Please note that" and "kindly" add seconds and no meaning.
Prompt writing checklist
- Read aloud once to catch awkward phrasing
- Each prompt under about ten seconds where possible
- Keys come after the label
- Dates, hours and numbers checked as spoken
- No internal names or jargon
- Someone outside the team has listened
Recording a human voice
A reliable recording process
- Prepare a scriptFinalise every prompt, including wrong-key and no-input messages.
- Choose the roomQuiet, soft furnishings, no echo, phone and fans off.
- Record in one sessionThe same voice, microphone and distance across all prompts.
- Edit gentlyTrim silence, even out volume, avoid heavy processing.
- Check on a phonePhone lines carry less audio than studios; verify clarity on a real call.
Choosing the speaker
Choose a clear, warm voice that suits your callers. A calm, medium pace works well. If you plan to have a synthetic voice take over some prompts, choose a recording voice that is not too dissimilar. If you use a voice other than your own, make sure you have the speaker's written permission to use it.
Using text-to-speech well
Modern synthetic voices can sound natural, but they still need care. Pick a voice for each language, listen to it on a phone line, and test with your actual text rather than a demo sentence.
Names, places and unusual words
Synthetic voices stumble on surnames, product names, place names and acronyms. Most systems let you supply a pronunciation or a reading for a word, and the effort is worth it for the five or ten terms callers hear most often. Where a language has several possible readings of the same written form, check each one in context.
Know the limits
- Long passages can sound flat. Break them up.
- Emphasis and emotion are limited; do not rely on tone to convey urgency.
- The same text may sound different after a voice update. Re-listen when a provider changes a voice.
- Hedge any claim of human-like quality. Callers may notice they are hearing a synthetic voice, and in some places you may have to say so.
The set of prompts every line needs
It is easy to write the greeting and forget the rest. Make a list of every message a caller might hear, so none is left as a default or a gap.
| Prompt | Purpose | Tip |
|---|---|---|
| Greeting | Say who you are | One short sentence |
| Main menu | Offer the choices | Label first, key second, pause between options |
| Wrong key or no input | Help the caller recover | Calm wording, repeat the choices once |
| Hold or queue message | Tell callers what is happening | Be honest about the wait; no endless promotions |
| Closed message | Give hours and a next step | Update when hours change |
| Voicemail prompt | Ask for name, number and reason | Short, with a clear tone to speak after |
| Goodbye | Close the call politely | Keep it to a few words |
Review the whole set together at least once. Prompts written on different days by different people often use different words for the same thing, such as "appointment" in one and "booking" in another, which makes the line sound disjointed.
Keeping audio clear and consistent
Audio qualities to check
- VolumeEven from prompt to prompt, with no sudden jumps.
- ClarityNo hiss, clicks or clipped words on a phone line.
- PaceSlow enough to follow, with natural pauses.
- ToneFriendly, calm and consistent with your brand.
Keeping prompts current
Treat prompts like website copy: they go out of date. Keep a simple log with the exact wording, the date, and the person who approved it. Review it each quarter and after changes in hours, services or staff. Avoid time-sensitive facts in recorded prompts where you can; put them in editable text instead. See Routing Calls by Business Hours, Holidays and Closures for how to keep hours-related messages accurate, and Text-to-Speech for Japanese Phone Systems for language-specific points.
Finally, keep the original text of every prompt somewhere safe, whichever method you use. If a recording is lost or a voice changes, you can recreate the audio quickly and keep the line sounding the same.
Doing it in Call
In Call, AI voices are managed under Voice. You can choose a voice for your workspace and give a single phone line its own voice. A pronunciation dictionary lets you teach the voice how to read names, places and words it gets wrong: you enter the word as written and a reading in kana, try a sentence to hear it, and save, with a limit of 100 words. A phone line's main language decides the language its calls start in. Voice cloning exists, with a consent step for the voice profile, so only clone a voice with the speaker's approval. The practice call and test call tools let you hear the result before callers do. See Choosing and Testing an AI Voice.
Frequently asked questions
Is a recorded voice better than text-to-speech?
Neither wins in every case. A recording sounds personal and suits stable greetings; text-to-speech is easier to update and suits changing information. Many teams use both.
How long should a prompt be?
As short as it can be while staying clear. Aim for about ten seconds for a menu, and split longer information into separate options.
How do I fix a mispronounced name?
Use the pronunciation settings of your voice system, if available, or rewrite the word phonetically and listen to the result on a phone.
Do I need a professional studio?
No. A quiet room, a decent microphone and a consistent setup are enough. Check the result on a real phone call.
Should I tell callers the voice is synthetic?
Practice and rules differ by country. Check your local regulator, and be honest if asked.
Can I use someone else's voice?
Only with their clear, written permission. This applies to recordings and to cloned voices.