Swap Voice

Two TTS backends, both configured in .config.yaml. For a curated list of voices that suit Dotty's persona (with character notes and best-for hints) see the Voice Catalog.

Piper (local, offline)

The fastest path is the install helper, which downloads any catalog voice into models/piper/ and (optionally) rewrites .config.yaml for you:

make voice-list                                       # see the catalog
make voice-install VOICE=en_US-kristin-medium APPLY=1
docker compose restart xiaozhi-server

Or do it by hand:

1. Download a voice .onnx + .onnx.json from Piper samples into models/piper/.

2. Update .config.yaml:

selected_module:
  TTS: LocalPiper
TTS:
  LocalPiper:
    model_path: models/piper/en_US-lessac-medium.onnx

1. Restart: docker compose restart xiaozhi-server

EdgeTTS (cloud, many voices)

1. List voices: pip install edge-tts && edge-tts --list-voices | grep en-
2. Update .config.yaml:

selected_module:
  TTS: EdgeTTS            # or StreamingEdgeTTS
TTS:
  EdgeTTS:
    voice: en-AU-WilliamNeural    # change to your pick

1. Restart: docker compose restart xiaozhi-server

Tips

• Piper is fully offline with no latency jitter. Prefer it for reliability.
• EdgeTTS has more variety but needs internet and occasionally throttles.
• English voices only -- non-English voices produce empty audio. See voice-pipeline.md.