One API for every local TTS model.

OpenAI-compatible speech. Named voices, saved locally.

GitHub

Quick start · Apache-2.0

pip install wavhost

Pullwavhost pull chatterbox-turbo

Runwavhost run chatterbox-turbo "Hello from your machine." -o hello.wav

Servewavhost serve:11435

Models (Chatterbox, Qwen3-TTS, Kokoro), voices, and cache live under ~/.wavhost. See Models. GPU recommended.

Named voices, saved locally

Create from short reference audio on the CLI or API, then reuse by name in run or POST /v1/audio/speech.

Createwavhost voice create my-voice --ref ./sample.wav

Listwavhost voice list

Runwavhost run chatterbox-turbo "Hello from Wavhost" --voice my-voice -o hello.wav

Storage: ~/.wavhost/voices/. Full CLI + API reference in Docs.

OpenAI-compatible API

Point a client at http://localhost:11435/v1. Pass a saved voice name, or "default".

curl http://localhost:11435/v1/audio/speech \
  -H "Content-Type: application/json" \
  -d '{
    "model": "chatterbox-turbo",
    "input": "Hello from Wavhost!",
    "voice": "my-voice"
  }' \
  --output speech.mp3

Voice CRUD: POST/GET/DELETE /v1/voices — see Docs.

Progressive download via stream: true for mp3 and pcm — see Docs.

Local & private

Runs on your machine — no API keys, no cloud account
Text, audio, and saved voices stay put
Model licenses printed on pull (Hugging Face)

What it isn't (v0)

Not cloud SaaS, marketplace, or voice sharing
Not speech-to-text or fine-tuning
Not a web voice studio — CLI + local API only