Skip to main content
What you get: an MP3 (or WAV / OGG) of your text read in the voice you pick, as a download URL. Input: up to 20,000 characters of text per request and a voice id. Split longer text into parts.

1. Install and set your key

2. Generate speech

Pick a voice

voice_id takes a catalog voice id (a number) or the id of a voice you cloned. List what your key can use:
Browse and preview the catalog in the voice library. To speak in your own voice, clone it first.

Direct the delivery

Directions go inline in the text:
  • Emotion and delivery: a leading bracket per segment, e.g. [whispering, tense].
  • Non-verbal sounds: [laugh], [sigh], [breathe], [chuckle], [gasp] and more.
  • Pauses: <break time="500ms"/> (up to 10 s each, 20 per request).
  • Pronunciation: IPA between slashes, e.g. Worcester /ˈwʊstər/.

Tips

  • speed runs from 0.25 to 4.0 (default 1.0). audio_format / audioFormat sets mp3, wav or ogg.
  • For a “listen to this page” button, call TTS from a server route so your API key never reaches the browser, then return output_url to the client.
  • Word-level timestamps for read-along UIs and designing a voice from a text description are covered in the Text to Speech reference.

From an AI agent

On the full MCP endpoint (https://mcp.audiopod.ai, see AudioPod for AI agents) ask: “Read this paragraph aloud in a calm voice: …”. The agent calls text_to_speech with text and, optionally, voice_id, language and speed.