Model / Voice Management

AbstractVoice core is remote-first:

• OpenAI remote audio (default): no local model download in the base install.
• Piper (local): small ONNX voices downloaded/managed by abstractvoice/adapters/tts_piper.py when you select tts_engine="piper".
• There is no legacy Coqui model management in core.
• Heavier engines (torch/transformers) are opt-in via extras (e.g. abstractvoice[chroma], abstractvoice[audiodit], abstractvoice[omnivoice]).

What gets downloaded, and when?

Piper voices are stored as small ONNX files under:

• ~/.piper/models

Downloads are controlled by allow_downloads:

• Library default: VoiceManager() uses OpenAI remote audio and does not download Piper models.
• Local Piper: VoiceManager(tts_engine="piper", allow_downloads=True) may download Piper models on-demand.

• REPL default: python -m abstractvoice cli runs with allow_downloads=False (offline-first), so it will not download implicitly.

• Typical voice size: tens of MB per language

Opt-in engines (HF cache)

Some optional engines download weights via Hugging Face and cache under ~/.cache/huggingface by default:

• Chroma cloning: python -m abstractvoice download --chroma (requires abstractvoice[chroma])
• AudioDiT (LongCat-AudioDiT-1B): python -m abstractvoice download --audiodit (requires abstractvoice[audiodit])
• OmniVoice: python -m abstractvoice download --omnivoice (requires abstractvoice[omnivoice])

In offline-first mode (allow_downloads=False) these engines will not fetch missing weights implicitly.

Programmatic introspection

• vm.list_available_models() returns a dict of known Piper voices and whether they are cached.
• vm.set_language("<lang>") loads the voice if cached; it will download only if allow_downloads=True.

CLI

• Use python -m abstractvoice cli and /voices models to view available Piper voices (and cache status). /setvoice still works as a compatibility command.
• Prefetch explicitly (offline-first):

python -m abstractvoice download --piper en