# Text to speech

Generate raw audio with model-defined voices, languages, and input limits.

`POST /v1/audio/speech` returns synthesized audio bytes rather than a JSON wrapper.

```bash
curl https://api.koscompute.com/v1/audio/speech \
  -H "Authorization: Bearer $KOSCOMPUTE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "kosmik/tts-kokoro-quality",
    "input": "The deployment completed successfully.",
    "voice": "af_heart",
    "speed": 1.0,
    "response_format": "wav"
  }' \
  --output speech.wav
```

## Parameters

`model` and `input` are required. `voice` defaults to the model's published default. `speed` is a numeric playback-rate control. Supported output formats are WAV, MP3, FLAC, Opus, and PCM; an unrecognized format currently falls back to WAV, so clients should send a documented value and check `Content-Type`.

The maximum input length, supported languages, voice IDs, and pricing are model metadata. The current public models use a configured per-request character limit; read `tts_max_chars` instead of duplicating that number in application code.

## Response

On success, read the response as binary audio. Relevant content types include `audio/wav`, `audio/mpeg`, `audio/flac`, and `audio/ogg` depending on the requested encoding. Save or stream the bytes directly.

Invalid input or voice selection returns `400`; oversized text returns `413`; capacity returns `429`; and no ready TTS route returns `503`. Honor `Retry-After` for temporary failures.