Status: final, publish-ready copy. Owner: mobi (t_293d061d). Synthesized from t_3cee5f4d (keyword/SERP), t_f0296baa (pricing), t_ce3a6c0d (verification, gate PASS). All vendor specifications and prices were read from first-party pages on 2026-10-30.
Target URL: /use-cases/ivr-voice-prompts Primary keyword: ivr prompt generator Secondary keywords: ivr audio format · 3cx voice prompt format · grandstream voice prompt package · freepbx text to speech Body length: 1,617 words of prose (H1 to final CTA; table rows and code samples excluded) · Reading level: Flesch-Kincaid 7.0, Flesch reading ease 69 (measured on the copy) Sections: 7 body sections + 9-question FAQ
Title tag (51 chars) IVR Prompt Generator for 3CX, FreePBX & Grandstream
Meta description (148 chars) Generate IVR prompts in the exact format your PBX accepts: 8 kHz mono WAV, G.711 µ-law, or a full import-ready prompt set. Preview, batch, download.
Canonical: https://[domain]/use-cases/ivr-voice-prompts
Internal links (anchor text)
/guides/ivr-audio-format ("IVR audio formats, per platform")/guides/freepbx-text-to-speech ("FreePBX text to speech setup")/guides/3cx-prompt-set ("building a 3CX prompt set")/pricing ("IVR prompt generator"), from posts about IVR scriptsIVR Prompt Generator: voice prompts your phone system actually accepts
Paste your call-flow script, pick your phone system, and download files it accepts on the first upload: 8 kHz mono PCM WAV for 3CX, G.711 µ-law for Asterisk and FreePBX, WAV or GSM for Grandstream UCM. No audio editor, and no renaming dozens of files by hand.
Primary CTA: Generate your first prompt Secondary CTA: See the format presets
Most AI voice tools hand you a 48 kHz stereo MP3 by default. That is not the file your PBX wants, and it will either reject it or play it back thin and quiet after an internal transcode. We start from the destination: you choose the phone system, and the export matches what its documentation asks for.
| Phone system | Format to upload | What the vendor documentation says |
|---|---|---|
| 3CX (IVR, queues, system prompts) | WAV, PCM, mono, 8 kHz sample rate, 16-bit, 128 kbps | 3CX lists exactly this for system prompts and requires the file and folder names in a prompt set to stay unchanged, including letter case. (3CX docs) |
| FreePBX / Asterisk | G.711 µ-law (Asterisk's native telephony codec), with A-law and sln also supported | Asterisk sound packages ship in sln, u-law, a-law and GSM. µ-law squeezes 14-bit audio into 8 bits, so it is converted from a 16-bit source rather than an 8-bit one. (Sangoma, VoIP-Info) |
| Grandstream UCM6xxx | 8 kHz WAV, GSM, Signed Linear, G.729, or uncompressed alaw/ulaw | File names must match the vendor's prompt list (ucm_voice_prompt_list.xlsx), the files go into a folder with a specific prefix, and the package includes info.txt. (Grandstream docs) |
| Other systems (Yeastar, Cisco, Twilio, hosted PBX) | Custom preset: set sample rate, channels, codec and container yourself | We do not publish specs we have not verified. Check your vendor's current documentation. The custom preset covers any sample rate from 8 kHz up. |
The IVR audio format decides whether the upload works at all, so start there. If you are not sure what your system wants, 8 kHz, 16-bit, mono WAV is the common denominator for on-premises telephony and the first thing to try.
FreePBX text to speech has the same problem one layer down: a TTS engine bolted onto Asterisk still has to write audio that Asterisk will play without transcoding on every call.
These are the failures we see most often. Where a vendor documents the rule, the quote and the page are linked. The rest is what the systems do in practice.
1. Wrong sample rate or bit depth. 3CX states 8 kHz, 16-bit, mono PCM for system prompts. A 44.1 or 48 kHz file is not that file, and the difference shows up as a prompt that plays badly or not at all.
2. Stereo instead of mono. A stereo prompt doubles the file size, and some systems downmix it poorly. Export mono unless your vendor says otherwise.
3. A renamed file inside a prompt set. 3CX requires the names and structure of prompt files and folders to stay exactly as shipped, down to letter case, and the prompt ID in the set's XML has to match the file name. Change either and the prompt will not appear in the Management Console.
4. Grandstream file names that do not match the prompt list. On UCM6xxx each file has to carry the name defined in ucm_voice_prompt_list.xlsx, and the whole folder is packaged together with info.txt.
5. Codec mismatch on Asterisk. If you have to transcode 16-bit linear WAV to µ-law on every playback, you pay for it twice: extra work on each call, and a lossy step on top of an already narrow band. Rendering µ-law directly skips both.
In all five cases the audio itself was fine and the packaging was what failed. That is the part a generic text-to-speech tool leaves to you.
Callers hang up on prompts that read "four million two hundred thousand dollars" for a balance, or rattle a nine-digit account number out as one enormous number. Prompt text has to be written for the ear, and variable data has to be marked for the engine.
What we do at generation time:
by digit with natural pauses, instead of as a single integer.
instead of numeric slashes.
prompt says them the same way.
you a click instead of an upload-test-retry cycle.
Example source line:
Your account ending in 4471 has a balance of $1,204.50, due on March 3rd.Rendered as: "Your account ending in <4> <4> <7> <1> has a balance of <one thousand two hundred four dollars and fifty cents>, due on <March third>."
A working phone menu is not one prompt. It is a welcome message, a menu with four options, a closed-office message, a wait message, a voicemail greeting, a holiday greeting, and the extensions that get re-recorded every quarter.
Paste the whole script, mark the segment breaks, and export a named set in one pass:
Welcome to Acme. Thank you for calling.
[cut: 01-welcome]
For sales, press 1. For support, press 2. For billing, press 3.
[cut: 02-main-menu]
Our office is currently closed. Please call back during business hours.
[cut: 03-closed]Each marker becomes the file name, in dial-plan order, inside a single archive. 3CX prompt-set mode goes one step further: it produces the folder and file structure the system expects, with names kept exactly as required, so nothing has to be renamed after download.
A Grandstream voice prompt package is the same idea with Grandstream's rules: the names from ucm_voice_prompt_list.xlsx, the prefixed folder, and info.txt in place before you upload.
Batch also means re-recording is cheap. Change the holiday dates, regenerate those three files, drop them in, keep the rest.
Translating a phone system is a naming and packaging problem before it is a translation problem. The moment a file name or a prompt ID drifts, the PBX quietly ignores that prompt.
The flow keeps the source prompt list as the single source of truth: import the vendor's English prompt list, fill in the translations, generate each prompt in the target language and voice, export with the original IDs and file names intact. One language or twelve, the package imports the same way.
Most voice tools sell credits, and credits hide the unit price. ElevenLabs publishes an API list rate of $0.08 per 1,000 characters. Its Creator plan costs $22 for 121,000 credits, which works out at $0.18 per 1,000: about 2.3 times its own list rate. You only see that by dividing.
Here are the published rates for the same nominal work, converted to one finished hour and read from vendor pages on 30 October 2026. The last column is the point: every vendor sells a different unit.
| Provider and plan | $ per finished hour | Sold as |
|---|---|---|
| Amazon Polly Standard; Google Standard/WaveNet | $0.24 | per million characters |
| Deepgram Aura-2; Murf Gen2; Amazon Polly Generative | $1.80 | per million characters |
| ElevenLabs API list rate (v4, v3) | $4.80 | per thousand characters |
| Amazon Polly Long-Form | $6.00 | per million characters |
| VocalLab | $7.20 | per finished minute |
| ElevenLabs Pro plan credits | $9.90 | credits |
| ElevenLabs Creator plan credits | $10.91 | credits |
| WellSaid Pro ($49 / 180 minutes) | $16.33 | per downloaded minute |
Finished hour = 60,000 characters, the same ratio ElevenLabs' own comparison table uses. Character-priced rates run from $4.00 to $181.82 per million characters, a 45.5x spread across the same nominal work. VocalLab and WellSaid sell minutes instead of characters, which is why they only join the same column after conversion.
We sell finished minutes, not credits. Regenerating a prompt is free, and a re-record is not a separate charge.
A two-minute check saves a support ticket:
sox --i prompt.wav or ffprobe prompt.wav.What audio format does 3CX need for IVR prompts? The 3CX voice prompt format is WAV, PCM, mono, 8 kHz sample rate, 16-bit, 128 kbps. 3CX publishes this for system prompts and also requires prompt file and folder names to stay exactly as shipped, including letter case.
What format does Grandstream UCM need? Grandstream lists 8 kHz WAV, GSM, Signed Linear, G.729 and uncompressed alaw/ulaw as compatible, and recommends GSM where file size matters. File names must match the ones in the vendor's prompt list, and the files are packaged in a folder with a required prefix plus info.txt.
What does Asterisk or FreePBX use? Asterisk's native telephony format is G.711: µ-law in North America, A-law in much of Europe, with signed linear (sln) and GSM also available. µ-law is 8-bit but is converted from 16-bit source audio for better quality.
Can I just use an MP3? Not for most on-premises PBX uploads. MP3 needs a decoder and is not the format the vendor documents expect. Use the preset for your system instead of converting after the fact.
How do I make my IVR read an account number digit by digit? Mark the digits for digit-by-digit reading at generation time, then preview. Reading a long number as one integer is the most common IVR audio mistake, and it cannot be fixed once the WAV is rendered.
Why does my prompt not show up after upload? On 3CX the usual cause is a prompt set whose file names, folder structure or XML prompt IDs no longer match the standard set. In that case the prompt is not used and does not appear in the Management Console. Import one prompt manually first to confirm the format, then load the set.
Do you support A-law for European phone systems? Yes. A-law is available alongside µ-law, and the custom preset lets you set sample rate and channels for systems whose documentation differs from the common defaults.
How many prompts can I generate at once? A full call-flow set goes out in one export: welcome, menu, queues, closed office, holiday, voicemail, each one named and ordered in a single archive. The size limit for a single export is shown in the app before you generate.
Is this a voice changer or a voice cloning tool? Neither. It generates new prompts from your script in a chosen voice, at the audio spec your PBX requires. No recording of real staff is needed.
Stop converting files after the fact. Generate the prompt in the format your phone system already wants, preview it, and upload it once.
[Generate your first prompt] · [See format presets]
{
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [
{"@type":"Question","name":"What audio format does 3CX need for IVR prompts?","acceptedAnswer":{"@type":"Answer","text":"The 3CX voice prompt format is WAV, PCM, mono, 8 kHz sample rate, 16-bit, 128 kbps. 3CX publishes this for system prompts and also requires prompt file and folder names to stay exactly as shipped, including letter case."}},
{"@type":"Question","name":"What format does Grandstream UCM need?","acceptedAnswer":{"@type":"Answer","text":"Grandstream lists 8 kHz WAV, GSM, Signed Linear, G.729 and uncompressed alaw/ulaw as compatible, and recommends GSM where file size matters. File names must match the ones in the vendor's prompt list, and the files are packaged in a folder with a required prefix plus info.txt."}},
{"@type":"Question","name":"What does Asterisk or FreePBX use?","acceptedAnswer":{"@type":"Answer","text":"Asterisk's native telephony format is G.711: µ-law in North America, A-law in much of Europe, with signed linear (sln) and GSM also available. µ-law is 8-bit but is converted from 16-bit source audio for better quality."}},
{"@type":"Question","name":"Can I just use an MP3?","acceptedAnswer":{"@type":"Answer","text":"Not for most on-premises PBX uploads. MP3 needs a decoder and is not the format the vendor documents expect. Use the preset for your system instead of converting after the fact."}},
{"@type":"Question","name":"How do I make my IVR read an account number digit by digit?","acceptedAnswer":{"@type":"Answer","text":"Mark the digits for digit-by-digit reading at generation time, then preview. Reading a long number as one integer is the most common IVR audio mistake, and it cannot be fixed once the WAV is rendered."}},
{"@type":"Question","name":"Why does my prompt not show up after upload?","acceptedAnswer":{"@type":"Answer","text":"On 3CX the usual cause is a prompt set whose file names, folder structure or XML prompt IDs no longer match the standard set. In that case the prompt is not used and does not appear in the Management Console. Import one prompt manually first to confirm the format, then load the set."}},
{"@type":"Question","name":"Do you support A-law for European phone systems?","acceptedAnswer":{"@type":"Answer","text":"Yes. A-law is available alongside µ-law, and the custom preset lets you set sample rate and channels for systems whose documentation differs from the common defaults."}},
{"@type":"Question","name":"How many prompts can I generate at once?","acceptedAnswer":{"@type":"Answer","text":"A full call-flow set goes out in one export: welcome, menu, queues, closed office, holiday, voicemail, each one named and ordered in a single archive. The size limit for a single export is shown in the app before you generate."}},
{"@type":"Question","name":"Is this a voice changer or a voice cloning tool?","acceptedAnswer":{"@type":"Answer","text":"Neither. It generates new prompts from your script in a chosen voice, at the audio spec your PBX requires. No recording of real staff is needed."}}
]
}re-record charge. These are offer commitments from the pricing analysis, not published facts. Confirm the product actually bills this way, or cut the last paragraph.
$0.05 to $0.10 per finished minute ($3 to $6 per finished hour), which sits about 1.7x to 3.3x the $1.80 per hour list rate of Deepgram Aura-2 and Murf Gen2, and below VocalLab ($7.20/hr), ElevenLabs Pro credits ($9.90/hr) and WellSaid Pro ($16.33/hr).
stable enough to publish.
[domain].vendor page is the wording used throughout.
synthesis-report.md §10. No customer counts, no "saves Xhours", no percentage claims anywhere in the copy.
tables. It beats narakeet by knowing the phone formats exist at all (narakeet scored 0 mentions of 8 kHz, 16-bit or mono in the page probe).