Record a voice sample for your custom AI voice
A custom voice is only as good as the recording behind it. Here is what the upload accepts, how to set up the room, and three short scripts to read aloud.
What the upload accepts
- Length
- 30 s – 5 min
- Of clear speech. The three scripts below take about 2½ minutes.
- File types
- MP3, WAV, WebM, OGG, M4A
- MP3 at 256 kbps or higher is the safe choice. WAV keeps more detail but passes 10 MB after about a minute.
- File size
- Up to 10 MB
- Two and a half minutes of 256 kbps MP3 is about 5 MB.
- Files
- One recording
- The upload takes a single file, so record the scripts in one take or join your takes first.
Set up the room before you press record
Do
- Record in a quiet room without echo. A small carpeted room or a closet full of clothes works well.
- Keep the microphone 15–30 cm (6–12 in) from your mouth, at the same distance the whole time.
- Speak in your normal voice, at your normal pace and volume.
- Use a headset or external microphone if you have one. A phone works, it just picks up more noise.
- Read each script straight through without stopping.
- Save as MP3 at 256 kbps or higher and 44.1 kHz. Use WAV only if the file stays under 10 MB.
Avoid
- Music, a TV, a fan or the air conditioning running in the background.
- Filters, EQ, noise gates, compression or voice effects. Send the raw recording.
- Anyone else talking, even briefly.
- Whispering, shouting or a put-on character voice, unless that is the voice you want.
- Recording over a phone or video call. The call compresses your voice before it is saved.
- Audio that clips, distorts or is barely audible.
Three scripts to read aloud
Read them in order, with a short pause between each. Together they cover the sounds a clone needs in about 2½ minutes.
- 0:00
Everyday narration
About 55 sA wide range of vowels and consonants at a natural pace.
Good morning. I wanted to take a moment to talk about how we spend our weekends. Some people like to wake up early and go for a long walk in the park, watching the sunlight move slowly through the trees. Others prefer to stay in, brew a pot of coffee, and read a book from cover to cover. Personally, I think the best weekends are the ones with no fixed plan at all, where you can decide, hour by hour, exactly what you feel like doing. Whether that means cooking a proper meal, calling an old friend, or simply doing nothing for a while, it’s the freedom to choose that makes it feel like a real break.
- 0:55
Questions and feeling
About 50 sQuestions, exclamations and a little emotional range.
Wait, seriously? You finished the entire project already? That’s incredible. I honestly didn’t think we’d be done until next week! I mean, don’t get me wrong, I’m thrilled, but how on earth did you manage it so fast? Was it the new schedule, or did the whole team just decide to push through together? Either way, congratulations. This is exactly the kind of result we’ve been hoping for, and I think everyone deserves a proper celebration tonight.
- 1:45
Numbers, names and terms
About 45 sDigits, times, spelled-out names and less common sounds: what callers hear most.
Please call me back at 555-201-4487 before 3:30 PM on Thursday, the 14th of August. My name is spelled R-I-L-E-Y, and the reference number for this order is TX-99204. We’ll need approximately twelve units, shipped from our warehouse in Zurich, and the total comes to $1,248.50 including tax. If anything changes, just email support@example.com and we’ll update the invoice.
- 2:30
Stop recording, trim the silence at each end and save.
Before you record
More questions? See the AI agents guides or contact us.
Can I record on my phone?
Yes. Use the built-in Voice Memos (iPhone) or Recorder (Android) app, hold the phone 15–30 cm from your mouth and record somewhere quiet. Don’t use speakerphone, and don’t record through a call.
Do I have to record everything in one sitting?
No, but the upload takes one file. Record each script as its own take if that’s easier, keep your tone, energy and microphone distance the same, then join the takes into one file before you upload.
What if I stumble mid-sentence?
Pause, then read the whole sentence again from the start. Trim the mistake out afterwards so the file you send has no stumbles, restarts or long “umm”s left in.
Does background noise really matter?
Yes. Fans, traffic and electrical hum get recorded along with your voice and carry into the clone, which can make it sound muffled or artificial. Use the quietest room you have.
Can I read something other than these scripts?
You can, as long as it lands between 30 seconds and 5 minutes and mixes statements, questions and numbers. The scripts here are written to cover a wide range of sounds in a short time, so they are the quickest route to a good result.
Can I clone someone else’s voice?
Only your own voice, or a voice you have explicit, documented permission to use. Cloning a voice without the speaker’s consent is not allowed.
Where do I upload the recording?
Custom voices are part of the Enterprise plan. Once custom voice is turned on for your account, open Agents in the app, choose Custom voice from your agent’s menu and upload the file. You can preview the voice before your agent uses it on calls.