Voice presets

The 56 built-in speakers Chorus exposes, and how to steer their delivery.

Overview

Chorus exposes 56 preset speakers. Pass an id as voice. Presets are the most consistent option, because the voice is fixed rather than generated per request.

voice accepts preset ids only. Any other value returns 400, with the model's available ids in the error message. Omitting voice does not fail: a request to chorus or chorus-clone with no voice and no reference audio synthesizes dominic, the cloning model's default speaker. Set voice explicitly if you do not want that fallback.

List the voices before hardcoding an id. The endpoint needs no API key.

curl "https://api.sprag.ai/v1/audio/voices?model=chorus"

Preset request

curl https://api.sprag.ai/v1/audio/speech \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $SPRAG_API_KEY" \
  -d '{
    "model": "chorus",
    "input": "The quarterly numbers are in and they look good.",
    "voice": "aiden",
    "response_format": "mp3"
  }' \
  --output out.mp3

Voices behind the `chorus` alias

chorus is a shared alias, and listing its voices unions the pools of the two models behind it. Nine are the built-in speakers of the preset model; the other 47 ship with the cloning model. All 56 report category: "preset" and all work the same way as a voice value.

To see which set an id belongs to, query that model directly.

curl "https://api.sprag.ai/v1/audio/voices?model=chorus-voices"

Preset-model speakers

Every preset reads any supported language. The language column names the accent the speaker carries, not a limit on what they read.

Voice id	Name	Character	Native language
`aiden`	Aiden	Sunny American male, clear midrange	English
`ryan`	Ryan	Dynamic male, strong rhythmic drive	English
`vivian`	Vivian	Bright, slightly edgy young female	Chinese
`serena`	Serena	Warm, gentle young female	Chinese
`uncle_fu`	Uncle Fu	Seasoned male, low and mellow	Chinese
`dylan`	Dylan	Youthful Beijing male, clear timbre	Chinese (Beijing)
`eric`	Eric	Lively Chengdu male, husky brightness	Chinese (Sichuan)
`ono_anna`	Ono Anna	Playful Japanese female, light and nimble	Japanese
`sohee`	Sohee	Warm Korean female, rich emotion	Korean

Cloning-model speakers

Another 47 presets take the same voice field.


`ali`	`amol`	`amrita`	`andal`
`arjuna`	`asan`	`balamani`	`bharathi`
`bulleh`	`dominic`	`fatima`	`haresh`
`iris`	`janaki`	`jayadeva`	`klaus`
`klaus_hessian`	`komal`	`krishna`	`kuntala`
`lachlan`	`lena`	`mahjoor`	`marin`
`miles`	`molla`	`narmad`	`padma`
`phulmani`	`priya`	`rahul`	`rajinder`
`rituparna`	`sahana`	`sanjay`	`sayali`
`sido`	`sienna`	`sneha`	`sourav`
`sunita`	`udit`	`vani`	`vemana`
`vince`	`wade`	`zooni`

The voices endpoint returns these with description: null. For accent, language, and a playable sample, use the voice library in the app. To compare two voices that read alike, synthesize a line of your own copy with each.

Steer delivery

Add instructions to a preset request to adjust how the speaker reads. The speaker itself does not change.

{
  "model": "chorus",
  "input": "Your table is ready.",
  "voice": "serena",
  "instructions": "Unhurried, slightly amused"
}

Voice design is a different operation on a different model, where instructions with no voice defines the speaker rather than modifying one.