Getting started

From install to first WAV and a named local voice. About five minutes if you already have a GPU-friendly Python env.

1. Install

pip install wavhost

This installs the runtime and PyTorch. TTS engines are pulled with the model — not bundled in the base package.

GPU tip

After pull, if PyTorch is CPU-only on an NVIDIA machine, Wavhost prints the exact pip install torch==…+cu124 line to run. Use the printed pin; don't invent a CUDA tag.

2. Pull a model

Chatterbox · MITQwen3-TTS · Apache-2.0Kokoro · Apache-2.0

wavhost pull chatterbox-turbo
wavhost pull chatterbox-multilingual
wavhost pull qwen-0.6-customvoice
wavhost pull kokoro

Shows the license, installs the engine if needed (chatterbox-tts, qwen-tts, or kokoro), and downloads weights into ~/.wavhost. See the full list on Models.

3. Generate speech

wavhost run chatterbox-turbo "Hello world, this is Wavhost!" -o output.wav
wavhost run chatterbox-multilingual "Bonjour, comment ça va?" -l fr -o fr.wav
wavhost run qwen-0.6-customvoice "Hello world" --voice Ryan -o qwen.wav
wavhost run kokoro "Hello from Kokoro" --voice af_heart -o hello.wav

Qwen CustomVoice speakers

Pass --voice Ryan (default), Aiden, Vivian, Serena, Uncle_Fu, Dylan, Eric, Ono_Anna, or Sohee. For cloning, pull a qwen-*-base model and pass a saved voice or audio path.

Kokoro named voices

Pass --voice af_heart (default) or any pack from wavhost show kokoro. Kokoro does not clone from reference audio.

4. Save a voice (optional)

wavhost voice create my-voice --ref reference.wav --desc "My custom voice"
wavhost run chatterbox-turbo "Hello from my voice!" --voice my-voice -o output.wav
wavhost run qwen-0.6-base "Hello from my voice!" --voice my-voice -o qwen-clone.wav

Voices live under ~/.wavhost/voices/. You can also create them over HTTP once the server is up — see API · Voices.

5. Start the API

wavhost serve

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

Default listen address is http://localhost:11435. Use "voice": "default" for the built-in stock voice (Qwen Ryan or Kokoro af_heart).