Voice Kit: voiceovers in your own voice or a stock voice

Voice Kit reads with a voice service. While that service is switched off, a yellow note at the top of the page says so, and every read uses a sample sound instead: soft tones paced like speech, one per syllable group, with real pauses at commas, periods, and [pause] marks. It lets you try every step and check the timing and length of a script, but it is not a voice. Wherever it appears it's labeled Sample sound, the stock voices are named Warm (sample) and Clear (sample), and MP3 download is off (the sample downloads as WAV).

Start a voiceover (left): pick the Language, then choose where the words come from with the four tabs. Type or paste: a box for the whole script (each paragraph becomes one line; the microphone button types what you say), then Start with this script, or Start blank, AI writes it to open an empty script and ask AI for one. A file: Word (.docx), PDF, text, Markdown, or HTML up to 8 MB. A web page: type an address and press Bring it in; the page's readable words come in. Writing Pack: a list of your Writing Pack drafts; click one to bring its words in. Very long sources come in up to 6,000 characters, and a note says so.

Your voiceovers lists every script you've made with its word count, rough length, language, and date; click one to open it (each has its own address, so you can bookmark or share the link with yourself). A search box appears once you have more than four.

Voices (right) shows your own voices and the stock voices. Pick who reads a script once a script is open.

The name at the top is the voiceover's name: click it to rename. Saved / Saving… beside it shows that every change saves on its own a moment after you stop typing.

Language sets the language the script is written and read in (English, Spanish, French, German, Portuguese, Italian, Dutch, Polish, Japanese, Chinese, Korean, Hindi). Length is the target: 15, 30, 60, or 90 seconds (press again to clear). On the right, the estimate (≈ seconds and words, at about 150 words a minute plus pauses) and, with a target picked, On target or how many seconds long or short.

Each line has its number, its words (click to edit, with a microphone button), and small buttons: the line's length with a play button once it's been read, Read this line again (after a first read), Line below, and the trash can to remove it. A line you change after a read gets an orange edge and the tag Changed: reads again. Edit as one block switches to a single box for the whole script (a blank line starts a new line); lines whose words you didn't touch keep their reads. Add a line, Add words from… (the same four sources, added at the end), and Delete (asks first, and removes the script's reads too) are at the bottom.

Write [pause] or [pause 1.5s] for a pause. If a name or word comes out wrong, spell it the way it sounds and read just that line again.

The violet box takes any request in your own words, typed or spoken: write the script, fit it to 30 seconds, translate it, add a pause after the phone number, mention your Saturday hours in line 2, make line 4 shorter. Press Make the change (or ⌘/Ctrl+Enter). The buttons under it are shortcuts, not the limit: Write it for me (on an empty script), Fit to the picked length, Make it friendlier, Easier to say, Stronger ending.

With Use my Brand Voice ticked (it shows when you've set one up in Brand Voice), the AI writes in it. Lines the request doesn't touch stay exactly as they were, so they keep their reads. The AI never makes up facts like prices or phone numbers: it writes a placeholder such as [your phone number] instead and lists it under Not done yet.

After each change a note lists what was Done and, first, anything Not done yet with the reason. Undo puts the script back as it was before that change.

Make your own voice (in the Voices panel) walks through three steps. 1, Your name: as you'll say it. 2, Read this sentence aloud: press Record the sentence, read the sentence on screen ("I, your name, agree to aiKit making an AI voice from my recordings, and I will only use it for my own work."), and press Stop. A red dot, the time, and a level meter show while it records. 3, 1 to 2 minutes of you talking: Record (or Upload a recording: MP3, M4A, WAV, WebM, or OGG), as many times as you like up to six recordings; a bar fills toward two minutes, at least 30 seconds is needed and a minute or more sounds best. Each recording is listed with its length and a trash can.

Make my voice runs the checks, then makes the voice. The consent sentence is turned into text and compared with the sentence (your name is left out, since names are spelled many ways); at least 70% of its words have to be heard. A basic same-voice check compares the pitch and tone of the consent recording with your samples; they have to match, so the sentence has to be read by the same person as the samples. It's a basic check, not a fingerprint. The result lists each check with a tick or a cross; if one fails, the voice isn't made and Fix the recordings takes you back.

Your recordings and voices are private to your account. The trash can beside a voice asks first, then deletes it at the voice service too, and here. Voiceovers already read with it stay yours. Up to five voices per account. Only your own voice: recordings of other people aren't allowed. Making your own voice is open to admins until a final legal review; stock voices work for everyone now.

Who reads it lists your ready voices first, then the stock voices (six at first; All stock voices shows the rest). The picked voice has a colored outline, and it's kept in the page address.

Hear it: the big button says what it will do: Read it with <voice>, Read the N changed lines, or Read it all again. Only lines that are new or changed are read again, so fixing one word costs one line. While it works, the panel says how many lines it's reading; long scripts are read in rounds and the count updates. If a line couldn't be read, Not done yet says how many and why, and pressing the button again reads just those.

Once every line is read, the player shows a waveform of the whole voiceover: press the round button to play or pause, click the waveform (or use the arrow keys on it) to jump, and the line being spoken shows underneath. Lines are joined with a short breath of silence between them.

Downloads and more: MP3 (real voices), WAV (any read), Save to Graphics Library (in the folder Voice Kit, category Voiceover), and Use in Motion Maker, which saves it to your library and opens Motion Maker with the voice on the Music tab, the music dipping under it. Under that, rate the read with a thumbs up or down; a thumbs down asks what went wrong and goes to the aiKit team. The AI voice / Sample sound tag at the top of the panel says which you're hearing.

Before new lines are read, a quick AI check reads them for impersonation: a script that pretends to be spoken by a real person or a real company that isn't you (for example a celebrity endorsing you in the first person, or "this is your bank calling") isn't read, and the panel says why. Mentioning or reviewing real people and companies, ads for your own business, and fictional characters are fine. Each line is checked once; changing it checks it again.

Voice Kit uses your account's AI credits at the voice service's real cost: reading is charged by the character (about 3 to 9 cents for a 30-second read, depending on the voice service), plus the small cost of each AI script change and safety check. Reading only the changed lines keeps it low. If a request would go past what's left on your plan, it isn't started and the message says so. Admins see each read's cost in cents under the player.

Admins see Side-by-side engine test under the player. With both voice services switched on, Read on both reads the open script with the picked voice on each service, shows a player and the cost for each, and Use this engine makes that one read every voiceover from then on (your own voice is made on both services when you make it). With one or none switched on, it says which are on.

Other aiKit tools ask Voice Kit for narration the same way: the words go in, and a voiceover comes back with each line's start and end time, so captions can follow the voice. It's read with your newest voice (or the first stock voice), counted on your AI credits, and marked as the sample sound whenever real voices are switched off.