LFM3 Audio for NVDA

Add-on and native host: Copyright 2026 Andre Louis, GPL-2.0-or-later.
NVDA playback, settings and sequence handling derive from Phonon for NVDA.
Download-state management follows Samsung Galaxy Voices' persistent background
job design. These are separate products; neither is required at runtime.

llama.cpp b12014: Copyright Georgi Gerganov and contributors, MIT.
https://github.com/ggml-org/llama.cpp/tree/b12014
The bundled runtime is the official Windows Vulkan x64 build. Selected CPU
backends and LLVM OpenMP are retained for compatible processor dispatch.
The native host uses llama.cpp's C APIs, trained TTS prompt and generation order.

Liquid AI LFM3-Audio-2B model: separately downloaded, LFM Open License 1.0.
https://huggingface.co/LiquidAI/LFM3-Audio-2B-GGUF
Tested model revision: 02355c1a141e464a064f2ab2c276638dbcb7f38a.
No model weights or voice recordings are included in this add-on.

NV Access provides NVDA and its audio APIs. Sonic is provided by NVDA and uses
the Apache License 2.0. NVDA's app-local Microsoft runtime is used by the helper;
Microsoft runtime files are not redistributed in this package.
