Speech

Text to speech, free every day

10,000 characters free on every account, every day, then $3 per million characters. Natural voices in 31 languages, and your own voice cloned from a short recording. Transcription too, from the same key.

OpenAI-compatible: client.audio.speech.create() works unchanged
mp3, wav, flac, opus or pcm, at 0.5x to 2x speed
Try every voice in the dashboard before you write code

Try it free API docs

One request, one audio file

curl https://api.sparrowcodes.com/v1/audio/speech \
  -H "Authorization: Bearer $SPARROW_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "supertone/supertonic-3", "voice": "F1",
       "language": "en", "input": "Your order has shipped."}' \
  --output speech.mp3

The response is the audio file. Headers show the characters used and how many free ones are left today.

Voices

Two speech models, one price

Both models share the daily free allowance and cost the same after it. Pick by language and whether you need a cloned voice.

Supertonic 3

$3 / 1M chars

Fast, natural text to speech in 31 languages with 10 preset voices (5 female, 5 male). Every voice speaks every language. Supports inline expression tags such as <laugh>, <breath> and <sigh>.

Languages: 31
Free a day: 10,000
Per request: 5,000 chars

Model details

Pocket TTS

$3 / 1M chars

Clone any voice from a 10 to 30 second recording, then use it for as much text as you like. Comes with 27 ready-made voices in English, French, German, Spanish, Italian, Portuguese and Dutch.

Languages: 7
Free a day: 10,000
Per request: 5,000 chars

Model details

Built for apps

Everything you need to ship a voice feature

Free every day: 10,000 characters a day on every account, shared across both models. Enough for prototypes and small apps.
Voice cloning: Record or upload 10 to 30 seconds of a voice in the dashboard, then use it like any other voice.
Many languages: 31 languages across the two models. Every Supertonic voice speaks every one of its languages.
The OpenAI format: Send model, voice and input to /v1/audio/speech. Existing OpenAI code needs only a new base URL.
Formats and speed: mp3, wav, flac, opus or pcm with response_format, and speed from 0.5 to 2.
Clear costs: Each response reports the characters used and the free characters left today in its headers.

Transcription

Speech to text from $0.003 a minute

Send an audio file to /v1/audio/transcriptions, the same multipart request as OpenAI's. Files up to 25 MB, billed by the second.

curl https://api.sparrowcodes.com/v1/audio/transcriptions \
  -H "Authorization: Bearer $SPARROW_API_KEY" \
  -F model=openai/gpt-transcribe \
  -F file=@meeting.mp3

Model	Per minute
GPT Transcribe OpenAI's recommended model for transcribing recorded speech in its original language.	$0.0045
GPT-4o Transcribe Accurate transcription from the GPT-4o family.	$0.006
GPT-4o mini Transcribe Lower-cost transcription from the GPT-4o family.	$0.003
Whisper OpenAI's original Whisper. Returns segment timestamps with response_format=verbose_json.	$0.006

Keep going

Make your first audio file

Start free Read the docs