How to Make a Character Voiceover With Text to Speech
How to make a character voiceover from a script, with numbers from our test: one 43-word line read by 14 character voices, six ways to write a pause, and 16 pitch and effect settings checked for lost words.
By Facecam team
Published · 7 min read

On this page
- What do you need to make a character voiceover?
- How do you make one, step by step?
- Which voice fits a fixed length?
- How do you add a pause that actually works?
- Can you make the character whisper or shout?
- Do pitch and effects make the words harder to understand?
- How long does each line take?
- What didn't we test?
- Questions
Short answer
Write the script, pick a character voice, generate it, then add pitch or an effect and download the WAV. In our test the voice choice alone changed the same 43-word line from 14.9 to 17.3 seconds. Ellipses and break tags didn't lengthen a pause, but the audio tag [long pause] did, from 0.46 to 1.72 seconds.
Every guide we read on 1 November gives the same four steps: script, voice, speed, export. Most add that punctuation controls the pacing. We tested that claim along with the rest, because a voiceover that runs two seconds long or skips a dramatic pause is the usual reason people regenerate a line ten times.
What do you need to make a character voiceover?
You need a script, a text-to-speech tool with character voices, and somewhere to place the audio, such as a video editor, OBS or a game engine. Character Voice Generator is a browser tool that reads your text aloud in an original character voice, with pitch and voice effects, and lets you download a WAV.
- Free, no account: 3 lines a day of up to 200 characters. Downloads end with a 2.28-second spoken Facecam tag.
- On a plan: up to 5,000 characters a line, at 13 credits per 1,000 characters, with no tag.
- Settings: 14 character voices, speed from 0.5× to 2×, pitch from -12 to +12 semitones, and 12 effects, from Robot and Villain to Radio and Cave Echo.
How do you make one, step by step?
Write for the ear
Short sentences, numbers spelled the way they should be said, and one idea per line. Read it aloud once; anything you trip over, the voice will too.
Open Character Voice Generator and paste the line
Free lines stop at 200 characters; the counter next to the button shows how many you have used. Longer scripts need a plan, or split the script into scenes.
Pick the voice by character and by pace
Each voice has a short description in the menu ("Older storyteller", "Low and husky"). If the voiceover has to fit a fixed length, check the pace table below too.
Add audio tags where the delivery matters
Put [long pause], [whispers] or [shouts] in square brackets right before the words. They aren't read out.
Generate, then set pitch and effect
Click Generate. Then change pitch or effect and click Apply pitch and effect: it reworks the same line in your browser and doesn't use a free line or credits.
Download WAV and place it
Drop it on your video timeline, into an OBS media source, or into your game's audio folder.
Which voice fits a fixed length?
The fastest voices read the same script about 15% quicker than the slowest. Our 43-word line ran 14.9 seconds in Kestrel and 17.3 seconds in Thistle, a 2.4-second gap. On a 30-second ad, that is the difference between fitting and not. As a rule of thumb from these numbers, a minute of voiceover is roughly 820 to 950 characters of script.
| Voice | Sound | Seconds | Words a minute |
|---|---|---|---|
| Kestrel | Calm and neutral, American | 14.9 | 173 |
| Atlas | Deep and casual, American | 15.0 | 172 |
| Saffron | Upbeat and clear, American | 15.4 | 167 |
| Marigold | Lively and expressive, American | 15.6 | 166 |
| Basil | Young and friendly, American | 15.6 | 166 |
| Coral | Bright and expressive, American | 15.7 | 165 |
| Orion | Laid-back and confident, American | 15.9 | 163 |
| Pixel | Soft and young, American | 15.9 | 162 |
| Clover | Crisp and confident, British | 16.0 | 161 |
| Rowan | Rich and authoritative, British | 16.4 | 158 |
| Juniper | Warm and friendly, American | 16.6 | 156 |
| Wren | Warm and velvety, British | 16.8 | 154 |
| Flint | Low and husky | 16.9 | 153 |
| Thistle | Older storyteller, American | 17.3 | 150 |
Pick the voice for pace before you touch the speed slider. Speed is applied after the voice is made, as a tempo change that keeps the pitch; we didn't test how far it can go before it sounds stretched, so we'd keep it between 0.9× and 1.1× and let the voice do the rest.
How do you add a pause that actually works?
Use the audio tag [long pause]. In the Thistle voice, the gap before "Gone." in "and the bridge? Gone." was 0.46 seconds as written. Every trick the guides suggest left it about the same, and the tag nearly quadrupled it.
| What we typed before "Gone." | Pause |
|---|---|
| Nothing extra ("bridge? Gone.") | 0.46 s |
| Two ellipses ("... ...") | 0.54 s |
| An em dash (the long dash) | 0.45 s |
| A blank line | 0.45 s |
| <break time="1.5s" /> | 0.45 s |
| [long pause] | 1.72 s |
The break tag result matches ElevenLabs' own docs, which say Eleven v4 doesn't support SSML break tags. Run on facecam.ai as a free visitor, the same tag gave a 1.81-second pause. We also stripped every comma, full stop and question mark from the full script and ran it in three voices. The model put its pauses in the same places anyway, and the length moved by at most 0.7 seconds. Punctuation helps the model read your meaning, but it won't stretch a pause.
Thistle, as written: 0.46 s pause
Thistle, with [long pause]: 1.72 s
Can you make the character whisper or shout?
Yes: put [whispers] or [shouts] before the line. With [whispers], none of the line was voiced (0% of the speech had a pitch, against 57% without the tag), so it was a real whisper and not just a quieter read. With [shouts], the median pitch went from 181 to 292 Hz and the line came out 4 dB louder. Writing words in capitals did much less: pitch rose from 181 to 194 Hz and loudness by under 1 dB.
"Keep your voice down…" as written
The same line with [whispers]
One tag went too far for a machine listener: [sad] made the line slow and breathy, and Whisper returned no words at all, though ElevenLabs Scribe got every word. If your voiceover will be auto-captioned, check the captions after an emotion tag.
Do pitch and effects make the words harder to understand?
Barely, in our test. We ran the Juniper version of the script through 16 settings and through two speech recognizers. ElevenLabs Scribe got all 43 words in 15 of them. The exception was pitch -12, where both recognizers heard "bridge" as "branch". Whisper also missed one word on Telephone and skipped the first sentence on Demon, which Scribe heard in full.
| Setting | Words lost | Length change |
|---|---|---|
| Pitch -5 or +5 | 0 | none |
| Pitch +12 | 0 | none |
| Pitch -12 | 1 ("bridge" heard as "branch") | none |
| Robot, Radio, Megaphone, Chipmunk | 0 | none |
| Telephone | 1 (Whisper only) | none |
| Villain, Monster, Alien, Ghost, Announcer, Cave Echo | 0 | +1.2 s tail |
| Demon | 10 (Whisper only; Scribe 0) | +1.2 s tail |
Watch the tail instead. Seven of the 12 effects add an echo or reverb that rings on for 1.2 seconds after the last word, so the file grows from 16.9 to about 18 seconds. Trim it in your editor if the next line has to come in tight.
Juniper, no effect (first 9 s)
Juniper with Villain (first 9 s)
How long does each line take?
A few seconds. The speech model returned each line of our script in 3.6 to 5.1 seconds. On facecam.ai, from clicking Generate to a playing line took 8.5 and 9 seconds as a free visitor. Pitch and effects run in your browser: half of our 16 settings took under a second, and Ghost, the slowest, about 4 seconds.
If you would rather have a famous or recognizable voice than an original character, that's the AI Voice Generator. Our comparison of text to speech and speech to speech explains when to type a line and when to record it. To put a finished voiceover under a clip, see how to change the voice in a video.
What didn't we test?
- Listening panels. Our numbers come from timing, pitch analysis and speech recognizers, not from people rating the voices.
- Every tag. We tried [long pause], [whispers], [shouts] and [sad]. ElevenLabs lists more; some may work less well in some voices.
- One script. A different text, or another language, will change the times in the table. The gap between fast and slow voices should hold.
- Lip sync and long scripts. We didn't time a full multi-minute script or sync a voiceover to an animated mouth.
Questions
How many characters of script make a minute of voiceover?
About 820 to 950 characters of English, going by our test, where a 237-character script ran 14.9 to 17.3 seconds depending on the voice. Slower storyteller voices sit at the low end.
Do SSML break tags work in AI voiceovers?
Not in ElevenLabs Eleven v4, which Facecam's Character Voice Generator uses: a <break time="1.5s" /> tag left the pause at 0.45 seconds in our test, and ElevenLabs' docs say v4 doesn't support them. The audio tag [long pause] gave 1.72 seconds.
Can I change the pitch without paying again?
Yes. In Character Voice Generator, changing pitch or effect and clicking Apply pitch and effect reworks the last line in your browser. It doesn't use a free line or credits.
Is the character voiceover free to use?
You can make 3 lines a day of up to 200 characters with no account. Free downloads end with a short spoken Facecam tag; lines made on a plan don't have it.
Keep reading

Voice
How to Do a Movie Trailer Voice: Pace, Pauses and Pitch
How to do a movie trailer voice with your own mic: how to read the line, then what a free preset and an AI narrator voice add, measured on the same take.
8 min read

Voice
How to Do NPC Voices as a DM (Without Wrecking Your Throat)
How to do NPC voices for a whole session: what to change yourself, what a voice changer can carry for you, and what we measured when we ran nine fantasy voices through a call codec.
7 min read

Voice
How to Find the Key of a Song You're Singing
How to find the key of a song by ear or with a free key finder, and what we learned running three AI-made vocals through one: how long to sing, and why the answer is sometimes the 'wrong' key with the right notes.
7 min read
Try it
Character Voice Generator
Character Voice Generator is a browser tool that reads your text aloud in an original character voice, with pitch and voice effects, and lets you download a WAV.
Open Character Voice Generator