Creating your voice

Where: Proxies ▸ Voices ▸ Add Voice Your voice reads every script in your videos. You make it once, from a recording of yourself, in your own account. This page is about making it. Choosing which voice speaks which language, trying voices out and tidying up is in The voice studio.

​

Three ways to add a voice

Press Add Voice. The first step asks which speech engine to use, and how to add the voice. If you are not sure which engine, take the one marked default. An instant clone is the right starting point. You can always make a professional one later. Pro clone shows not available on engines that don’t offer it. The second step sets the language. Your recording should be in that language: the recording script you are given follows it.

​

Making an instant clone

In the third step, press Record with Script to record in the browser, or Browse Files to upload a recording you already have (up to 10 MB). A few tips appear before you start. Read the script on screen, press Stop when you are done, give the voice a name and press Create Voice Clone. It is usually ready within a minute, in the voice studio.

​

Making a professional clone

A professional clone learns from much more of your voice, so it takes longer. You can do it two ways:

• Guided recording. The script comes in blocks of about five minutes of speech each. Record them one at a time with Record this block; you can listen back and Re-record any block. You can also download the script as a PDF to read from.
• Upload files. Upload recordings you already have, for example from a studio. See Uploading studio recordings below.

A counter shows how much audio you have against the 30-minute minimum. When it is reached, name the voice and press Start training. Training takes up to three hours and uses one of your account’s professional-clone slots; the card in the voice studio updates by itself when it is done. The recordings and the language cannot be changed once training starts.

​

Recording well

The recording is the input, so its quality is your voice’s ceiling.

• A quiet room. No open window, no fan, no café. Soft furnishings help more than equipment.
• Your usual microphone is fine. A recent laptop’s built-in microphone often records better than a phone or earbuds. What matters more is the room, and keeping the same distance. If you are unsure, make a short test and listen back.
• Speak at your normal presenting volume, as if to a customer across the table. Don’t speak softly: the browser turns a quiet voice up for you, so it looks fine while you record, but your voice learns the soft, uneven delivery.
• Read at your normal pace, with a short pause between sentences. Not a newsreader, not a rush.
• Keep the same distance from the microphone throughout.
• Read the whole script you are given. It is written to cover the sounds the system needs.
• Listen back before you accept it. If you would not want a customer to hear it, record again.

​

We check every recording

Every recording and every uploaded file is checked before it is used, because a voice can only be as good as what it learned from. The check looks at four things:

• Loud enough. A quiet recording makes a quiet voice, and turning it up later brings the room noise up with it.
• A quiet background. Noise between your sentences ends up in every video.
• Not distorted. A recording that hit the maximum level (a microphone set too loud, or too close) cannot be repaired.
• Long enough. At least 10 seconds of speech.

When you record in the app, the check runs the moment you press Stop. If the recording is not good enough, you see why, with a tip, and record again straight away. If your volume went up and down a lot, you get advice to record again at a steady volume. You can also keep it: the voice will still work. When you upload a file, the check runs as soon as you add it: Files you chose to use anyway stay marked in the list, so you can replace them later.

​

Uploading studio recordings

For a professional clone you can upload recordings from a studio instead of recording in the app. Ask for:

• Level: about −20 LUFS integrated (between −23 and −18), peaks no higher than −3 dB. If in doubt, normalize to −20 LUFS before exporting. Quiet but clean files are raised automatically.
• Raw audio: no compression, noise reduction or EQ. The voice learns every effect.
• The pauses kept in: don’t cut out the silences between sentences. They are where we measure the background noise, and where the voice learns natural pacing.
• One voice, one room: a steady distance to the microphone and a quiet room. WAV, or MP3 at 192 kbps or more.

No studio? Recording in the app usually gives the cleanest result: the browser keeps your level steady for you.

​

Next