Voice samples

Generate a spoken sample of any voice from your own text.

Voice samples turn a short piece of your own text into speech in any ThunderPhone voice — any catalog voice or any of your organization's ready custom voices. Use them to audition voices, preview a greeting before an agent goes live, or let your users hear a voice say their own words.

Samples are meant for exactly that — sampling. Text is capped at 500 characters and generation is rate limited per organization. If you need longer samples or a higher limit, contact support@thunderphone.com.

Endpoints

Method	Path	Description
`POST`	`/v1/voices/preview`	Generate a sample and get a signed audio URL
`GET`	`/v1/voices/previews/{id}`	Stream a generated sample's MP3

Generate a sample

cURL

curl -X POST https://api.thunderphone.com/v1/voices/preview \
  -H "Authorization: Bearer sk_live_YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "voice": "sarah",
    "language": "en",
    "text": "Thanks for calling Northwind Dental. How can I help you today?"
  }'

Field	Type	Required	Description
`voice`	string	yes	A voice `name` from `GET /v1/voices` — a catalog voice like `sarah`, or one of your ready custom voices like `custom:cv_2f6f90b0e9a34ee8b39be7d1`
`language`	string	yes	Language code for the sample. For a custom voice, this must be the clone's language
`text`	string	yes	The text to speak, 1–500 characters

Response (200 OK)

{
  "id": "vp_9c2f90b0e9a34ee8b39be7d1",
  "url": "https://api.thunderphone.com/v1/voices/previews/vp_9c2f90b0e9a34ee8b39be7d1?sig=...",
  "cached": false,
  "charged_cents": 3,
  "expires_at": "2026-08-07T14:12:08.317Z"
}

Field	Type	Description
`id`	string	Identifier of the generated sample
`url`	string	Signed URL that streams the sample's MP3 without authentication
`cached`	boolean	`true` when the sample was served from cache instead of freshly generated
`charged_cents`	integer	What this request cost: `3` for a fresh generation, `0` for a cache hit
`expires_at`	timestamp	ISO 8601 time the signed URL stops working

Pricing and caching

A fresh generation costs 3¢ from your prepaid balance. Repeating an identical request — same voice, language, and text — is served from cache, costs nothing, and returns "cached": true with a fresh signed URL. That also means expired URLs are cheap to replace: re-send the same request and you get a new one free.

Generation is limited to 30 fresh samples per organization per hour; cache hits don't count against the limit.

Fetch the audio

The url in the response is a signed link to:

GET /v1/voices/previews/{id}?sig={signature}

It streams audio/mpeg without an Authorization header, so you can hand it straight to an <audio> element or a media player. Signed URLs are valid for 7 days. You can also call the endpoint without sig while authenticated:

cURL

curl \
  https://api.thunderphone.com/v1/voices/previews/vp_9c2f90b0e9a34ee8b39be7d1 \
  -H "Authorization: Bearer sk_live_YOUR_API_KEY" \
  --output sample.mp3

An unsigned, unauthenticated request — or one with an invalid or expired signature — returns 401 Unauthorized. A sample generated for another organization's custom voice returns 404 Not Found.

Errors

Status	Code or response	When
`400`	`{"voice":"Voice is required."}`	Voice is missing
`400`	`{"language":"Language is required."}`	Language is missing
`400`	`{"text":"Text must be between 1 and 500 characters."}`	Text is empty or too long
`400`	`{"voice":"Voice is not available."}`	Custom voice doesn't exist, isn't ready, or belongs to another organization
`400`	`{"language":"Language is not available for this voice."}`	The voice doesn't speak that language
`402`	`insufficient_balance`	Prepaid balance is below the sample price
`422`	`preview_unavailable_language`	Sample generation isn't available for this voice in that language yet
`429`	`preview_rate_limited`	The organization used its 30 fresh generations this hour

The 402, 422, and 429 responses share one shape:

{
  "code": "preview_rate_limited",
  "detail": "Sample generation limit reached. Try again later."
}

A 422 means the voice exists but ThunderPhone can't generate ad-hoc samples for it in that language yet — the voice still works normally on calls. Pick another voice or language for the sample.

Voice samples

Endpoints#

Generate a sample#

Pricing and caching#

Fetch the audio#

Errors#