Text to baa.

Type anything. Pick a lamb. Get back the sound of sheep, as long as your text takes to read aloud. In every language, with no exceptions and no words.

As long as reading it aloud

People read aloud at about 150 words a minute. Words are a poor unit across languages (Finnish fits a sentence into one word), so we count letters and digits instead: 150 words of about five letters each is 12.5 letters a second. Spaces and punctuation are free. Anything you type gets at least one second of sheep.

Then we pick baas at random, play each a little higher or lower depending on the voice, leave a pause after each, and pad the end with silence so the file lasts exactly the estimated time, to within 26 milliseconds.

One endpoint, familiar shape

If you have used a text-to-speech API, you already know this one. Any xi-api-key works; baa-demo-key is traditional.

curl -X POST "https://YOUR-HOST/v1/text-to-speech/dolly" \
  -H "xi-api-key: baa-demo-key" \
  -H "Content-Type: application/json" \
  -d '{"text": "Hello world", "model_id": "baa_multilingual_v1"}' \
  --output baa.mp3

• Path: POST /v1/text-to-speech/{voice_id}. Optional query output_format: any mp3_ value is accepted; you get 44.1 kHz mono MP3 at 64 kbps whatever you ask for.
• Body: text (required, up to 5000 characters), model_id (baa_multilingual_v1 or baa_flash_v1), seed (integer, same seed, same baas). Other fields are politely ignored.
• Answer: audio/mpeg, with headers character-cost, request-id, x-baa-estimated-seconds and x-baa-duration-seconds.
• Errors: JSON with a detail: 401 invalid_api_key when the header is missing, 404 voice_not_found, 422 for blank or overlong text, an unknown model, or a format that is not MP3.

Voices

• dolly Dolly. Warm, classic ewe. The default.
• lambert Lambert. A young lamb. High, earnest, a little breathless.
• ewen Ewen. A ram with a baritone. Audiobooks and late night radio.
• baarbara Baarbara. Newsreader. Crisp and slightly above it all.
• gary Gary. A human who bleats. Method actor, no sheep involved.

Free sheep

Every baa is cut from recordings on Wikimedia Commons that are in the public domain or dedicated to it under CC0. No attribution is required; we give it anyway.

• Sheep_bleat.ogg by Eviatar Bach. CC0 1.0.
• Sheep_bleating.ogg by earthcalling. Public domain (PDSounds).
• Corner_of_a_sheep_field_in_summer.ogg by earthcalling. Public domain (PDSounds 341).
• Man_bleats_like_goat_or.ogg by stilgar. Public domain (PDSounds).