> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.kotoba.tech/overview/capabilities/text-to-speech/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.kotoba.tech/_mcp/server.

# Text to Speech

Kotoba's Text-to-Speech (TTS) synthesizes audio from text and streams it
back as soon as the model produces it, so you can pipe straight to a
speaker, a WebRTC track, or an LLM-driven agent without waiting for the
full utterance.

Supported languages: English (`en`), Japanese (`ja`), Korean (`ko`),
Chinese (`zh`), and Spanish (`es`).

## Three call shapes

* **One-shot** — pass a full string in, get a complete waveform back.
* **Streaming** — send the text in one frame and read audio chunks as
  they emerge from the server (output streaming).
* **Async** — the same surface, with `AsyncKotobaClient` for
  production-grade concurrency.

Available Japanese speakers: `ja-man-m02-azawa` (male) and
`ja-woman-f04-me` (female).

## Where to go next

* [Python SDK for TTS](/t2s/python-sdk) — `client.tts.synthesize(...)`
  for one-shot, `client.tts.stream(...)` for streaming.
* [API reference](/t2s/streaming) — the AsyncAPI spec for the TTS WebSocket
  channel.

> Synthesize natural speech from text, with streaming output.