Choose a voice

Every text to speech request names a voice, and the voice has to belong to the model you ask for. This page shows where the voice ids come from and which ones each model takes.

Voice list: Live

How voices are listed#

GET https://api.fluxrouter.ai/v1/audio/voices returns the voices your key can use. It makes no speech and is not billed, and it works on any valid key, including the Free plan, so you can build a voice picker before you upgrade.

curl https://api.fluxrouter.ai/v1/audio/voices \
  -H "Authorization: Bearer $FLUX_API_KEY"

{
  "object": "list",
  "data": [
    {
      "id": "nova",
      "model": "flux-voice-speak",
      "name": "Nova",
      "gender": "neutral",
      "description": "Standard voice",
      "languages": ["en"],
      "entitlement": "paid_cleared"
    }
  ]
}

Use a row's model and id as the model and voice of a speech request.

• The list is filtered by your key. It shows only voices whose model your key may use. If a model is missing from your key's model list, its voices are missing here too. Read the list with the key that will make the speech.
• Send the key as Authorization: Bearer. This endpoint does not read an x-api-key header.
• Other fields may appear. Read the fields above and ignore anything else.

The full contract is in the endpoint reference.

The voices#

The catalogue has 61 voices in all. The table lists the voices on the two speech models these docs cover, read from the live list. Your key's own list is the final word.

Voice ids by model#

The voice list can include voices on other speech models too. Those models are not covered in these docs yet; if you use them, keep the model and id from the same row together.

A voice that does not belong to the model gets 400 with param: "voice", and the error message lists the voices that model takes.

On the Flux tools server, the speak tool takes a voice id on its own and picks the model that owns it.

Entitlement values#

Each row carries an entitlement that says whether your key can make speech with that voice right now. Every row in one response has the same value, because it depends on your plan, not on the voice.

Caching the list#

The list changes rarely. Cache it per key and ask for it again with the ETag you were given:

curl -i https://api.fluxrouter.ai/v1/audio/voices \
  -H "Authorization: Bearer $FLUX_API_KEY" \
  -H 'If-None-Match: "<etag from the last response>"'

When nothing has changed you get 304 Not Modified with no body, and you keep your copy.

• The response is marked Cache-Control: private, no-cache and Vary: Authorization. Shared caches do not store it, and your own cache should check it with the ETag before each use.
• Keep one cached copy per key, not one per app. Two keys can see different lists.

Do not call the speech endpoint to preview a voice for a "try it" button: every call is billed. Make one short sample per voice once, store it, and play that.

Next#

• Text to speech
• Text to speech problems