Voices
Spoken lines for characters, a narrator or an announcer, in 19 voices, with each word's timing for subtitles.
The Voices bench says a line of text out loud: a character's dialogue, a narrator, an announcer, a tutorial voice. Pick a voice, write the line, and hear it in the frame before you download it.
Picking a voice
The menu next to Settings lists 19 voices, ElevenLabs' premade set. The play button beside it says a sample line in the chosen voice. George is the default. Keep one voice per character across all of their lines.
| Voice | Sounds like |
|---|---|
| Adam | deep, firm, middle-aged American; announcer, narrator |
| Alice | clear, engaging British; tutorials, a guide |
| Bill | old, wise, crisp American; a mentor, a storyteller |
| Brian | deep, resonant American; trailer voice |
| Callum | husky trickster with an unsettling edge; villain |
| Charlie | young, confident Australian; hero, sidekick |
| Chris | charming, down-to-earth American; NPC chatter |
| Daniel | steady British broadcaster; announcer, commander |
| Eric | smooth, trustworthy American; shopkeeper |
| George | warm British storyteller; narrator |
| Jessica | playful, bright, young American; heroine, companion |
| Laura | enthusiastic, quirky, young American; sidekick |
| Liam | energetic young American; hype, tutorials |
| Lily | velvety British actress; queen, sorceress |
| Matilda | knowledgeable, professional American; guide, AI |
| River | relaxed, neutral American; narrator, computer |
| Roger | laid-back, casual American; NPC |
| Sarah | mature, reassuring American; mission control |
| Will | relaxed young American optimist; friendly NPC |
Writing the line
Write exactly what is said, the way it is said. Punctuation shapes the delivery: a full stop or a comma gives a pause, an ellipsis trails off, a question mark lifts the end. Write numbers and abbreviations as they should sound: "twenty-three", not "23".
Audio tags in square brackets direct the performance without being spoken:
[whispers] Keep your voice down. [pause] The guards change at midnight, and that's our only chance.[shouting] Fall back! Fall back to the gate! [panting] Go, go, go!Tags that work well: [whispers], [shouting], [laughs], [sighs],
[gasps], [excited], [sarcastic], [curious], [pause]. A line can hold
up to 5,000 characters, but one line of dialogue per job is easier to work
with: each can be redone on its own, and the game plays it at its own moment.
Settings
- Model: ElevenLabs v4 (the default) is the newest. v4 Turbo costs half as much. v3 is the previous expressive model.
- Delivery: Natural is the default. Expressive follows the audio tags most and varies most between takes; Steady keeps an even read.
- Language: Auto reads it from the text. Set it when a short line could pass for another language, such as a name or a single word.
- Seed (v4 and v4 Turbo): set one to get the same take back from the same line. Leave it empty for a new take every time.
Speech is priced by the character: 8 credits per 1,000 on v4, 4 on v4 Turbo and 10 on v3. A line under 150 characters costs 1 credit on any of them. The price shows under the prompt before you start.
Downloads
Every line comes as the same three audio files as music and sound effects (WAV, OGG and MP3), plus Word timings: a JSON file with each word's start and end in seconds.
{
"words": [
{ "word": "Keep", "start": 0.12, "end": 0.31 },
{ "word": "your", "start": 0.31, "end": 0.44 }
]
}Use it to show subtitles word by word, or to cut a long take into pieces. Show the line as a subtitle in the game too, since many players have the sound off.
API
The bench is the speech job kind. See Speech for the
parameters, and MCP for the generate_speech tool.