Spero · English text to speech
Send text, get speech back.
Spero is an English text-to-speech model with two US voices. You call it with one HTTP request and pay per character. Every new account gets $2 of credit.
Samples
Hear both voices.
Fictional lines rendered ahead of time by the same model the API serves. No real people or real calls.
Voices
Two voices, made with consent.
Both voices are built from recordings of real people who agreed to this use. Our terms forbid using them to impersonate anyone or to pass the audio off as a human recording.
- spero-f1
- Female, US English
- spero-m1
- Male, US English
API
One endpoint, the shape you know.
POST /v1/audio/speech takes the same request as OpenAI's speech endpoint, so its client libraries work if you change the base URL.
- model
- spero-v1
- voice
- spero-f1 or spero-m1
- input
- 1 to 2,000 characters
- response_format
- mp3, wav, flac, opus or pcm
Speed
Slow today, and we say so.
A two-sentence passage takes about 30 seconds, and a full 2,000 characters about 3 minutes. Each request renders six takes and keeps a clean one, and the audio arrives as one file, not a stream. That suits prerecorded audio, not live conversation. A faster engine is what we work on next.
Pricing
$20 per 1M characters.
Prepaid, with no subscription. You are charged when the render finishes, and a failed render is never charged. Each new account gets $2 of credit, or 100,000 characters, after confirming its e-mail.
Create an account- Input
- 2,000 characters a request
- Rate
- 60 requests a minute per key
- Concurrency
- 2 requests at once per key, and 2 per account
- Keys
- 10 active keys per account
- Weights
- Open-source, coming soon