Voice Design (MiniMax)

Generate a synthesized voice profile from a natural-language description. No reference recording needed. Returns a voice_id usable in /v1/audio/speech.

POST /v1/audio/voice-design

Voice Design (MiniMax)

cURL

curl --request POST \
  --url https://kymaapi.com/v1/audio/voice-design \
  --header 'Authorization: Bearer <token>' \
  --header 'Content-Type: application/json' \
  --data '
{
  "description": "<string>",
  "model": "<string>",
  "name": "<string>",
  "preview_text": "<string>",
  "gender": "<string>",
  "age_group": "<string>"
}
'

Synchronous endpoint. Describe a voice in plain English, get back a voice_id you can immediately use in /v1/audio/speech on any MiniMax voice model.

Use this when you don’t have voice talent, you’re prototyping a fictional character, or you want a brand-safe persona voice from scratch.

curl -X POST https://kymaapi.com/v1/audio/voice-design \
  -H "Authorization: Bearer $KYMA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "description": "Warm female narrator with a slight British accent, mid-30s, calm cadence",
    "gender": "female",
    "age_group": "young"
  }'

Request

application/json body.

description · string · required

Natural-language voice description. Max 1000 characters. Also accepts the alias text.

model · string · default: "minimax-voice-design"

Voice design SKU. Currently only minimax-voice-design is supported.

name · string

Optional human-readable label, max 64 chars.

preview_text · string

Optional sample text MiniMax will render in the new voice for an internal preview. Max 500 characters. Doesn’t appear in the response — the audio bytes are not returned (you’d call /v1/audio/speech afterward to render).

gender · string

Optional hint: male or female.

age_group · string

Optional hint: child, young, middle-aged, or elderly.

Response

200 OK JSON. Same shape as /v1/audio/voice-clone.

{
  "voice_id": "kyma_a91f4d2e7c8b5301",
  "name": null,
  "model": "minimax-voice-design",
  "cost_usd": 4.20,
  "balance_usd": 45.80
}

Pricing

Flat $4.20 per designed voice. One-time charge — once designed, the voice_id is reusable in unlimited TTS calls.

Voice design costs ~2× voice clone because synthesizing timbre from text is strictly more compute-intensive than reproducing a captured voice.

Ownership

Same gating as voice clone — designed voice IDs are owned by the requesting user. Sharing the voice_id with another account is rejected with 403 voice_not_owned.

Errors

Status	error.code	When
400	not_a_voice_design_model	model is not a design SKU
400	description_too_long	description > 1000 chars
400	invalid_request	missing description
402	insufficient_credits	balance below $4.20
500	ownership_write_failed	design succeeded but ownership row insert failed
502	provider_error	upstream MiniMax failure

See also

• POST /v1/audio/voice-clone — clone from reference audio instead
• POST /v1/audio/speech — use the designed voice
• Voice Design (model) — service overview