100+ languages · Unity / Unreal / Godot
Dialogue audio that builds like the rest of your game.
Acteme turns your script into mastered, correctly named voice files. Cast each character once, direct lines with style notes and vocal tags, render every take in one batch, and pull the results straight into your engine.
3,000 credits every month on the free plan — about 3.5 minutes of finished audio.
Varga
Studio voice · Algenib
wounded, trying to sound calm
I'm fine. <cough> It's just a scratch. <short pause> Keep driving.
HALO
Studio voice · Iapetus
gentle, a little lonely
Welcome back, Commander. You were gone for eleven years. I kept the lights on.
Borin
Acteme Special · Borin
warm, tired, good-natured
Forty years I've kept this store. <laugh> Never once had a customer pay full price.
Guard Captain
Studio voice · Alnilam
commanding, slightly out of breath
Halt! <short pause> Hands where I can see them.
How it works
Import the script, cast it once
Import CSV, JSON or Yarn Spinner. Line keys stay stable, so re-importing a revised sheet only touches what changed. Give each character a voice and a description, and every line they speak uses it.
- Studio voices
- 30 curated voices, from gravelly to breezy.
- Voice library
- Hundreds more, filterable by language, accent, gender and pitch.
- Acteme Specials
- Character voices our team built for games. Included on every plan.
| Varga | Algenib |
| HALO | Iapetus |
| Borin | Borin (Special) |
| Guard Captain | Alnilam |
- Style
- wounded, trying to sound calm
- Transcript
- I'm fine. <cough> It's just a scratch. <short pause> Keep driving.
Three unedited takes of the same line. Take 2 is the one we kept.
Direct each line, keep the best take
The model reads your text as a verbatim transcript. How a line is delivered goes in the style note: wounded, trying to sound calm. Things that happen at a precise moment — a cough, a gasp, a beat — go inline where they happen.
Write [angrily] in the text and the model will say “angrily”. The script linter flags that before you spend credits, along with tags the model doesn't know. Render up to ten takes per line and pick one; it stays picked.
Inline vocal tags
Two-speaker exchanges render in a single pass with natural turn-taking and backchannels (|mhm|) when both speakers use studio voices.
What it doesn't do, so you don't find out in week six.
- No runtime generation. Everything is rendered ahead of time. For live, per-player speech you want a realtime API, not an asset pipeline.
- Variables need variants. “You have {gold} gold” can't be voiced as one file. The Yarn importer warns you and you render the variants you need.
- No sound effects. Tags cover human vocalisations — breaths, laughs, grunts. Footsteps and gunfire stay in your SFX library.
Pricing
One credit per spoken character. No seats, no per-project fees. Cache hits, failed renders and cancelled items are refunded.
Prototype
$0/ month
Temp VO for a jam or a vertical slice.
- Projects
- 2
- Parallel renders
- 1
- Audio kept
- 15 days
- Queue
- Standard
- Ship commercially
- No
Indie
● for shipping$15/ month
One shipped game, full cast, a couple of languages.
- Projects
- 10
- Parallel renders
- 3
- Audio kept
- 3 months
- Queue
- Standard
- Ship commercially
- Yes
Studio
$45/ month
Multiple projects, localization passes, large batches.
- Projects
- 100
- Parallel renders
- 8
- Audio kept
- 6 months
- Queue
- Priority
- Ship commercially
- Yes
Need more in a crunch week? Top-ups from 100,000 credits for $12, valid 12 months. Every plan gets all voices, engine plugins and MCP.
Common questions
Can I ship the audio in a commercial game?
Yes, on Indie and Studio plans. The Prototype plan is for jams, prototypes and pitches. Some storefronts ask you to disclose AI-generated content; the manifest records the model and voice for every file so filling that form is easy.
Does this replace voice actors?
It shouldn't have to. Most teams use Acteme for prototypes, placeholder passes, barks and systemic lines, localization scratch tracks, and to hand actors a reference read.
What happens when a writer edits a line?
The line is marked stale. Render with scope “stale” and only those lines are regenerated. The engine plugin sees a new hash and replaces just that file, keeping the same path.
Is it deterministic?
Generation isn't bit-for-bit repeatable, which is why takes exist: render three, pick one, and it stays picked. Re-rendering the exact same inputs returns the cached take instead of charging you for a new one.
Can it do real-time dialogue for NPCs?
No. Acteme produces files ahead of time. That's deliberate: no runtime dependency, no latency, no per-player cost, and the audio passes through your normal mixing and QA.
Which languages work?
Over 100 languages, the same set on every model. The spoken language is detected from the text. You can machine-translate a script with tags preserved, then edit translations before rendering.
Import a script. Hear the first line in a minute.
Google sign-in, no card, free credits every month.