GHOST Community Edition v0.1.0

GHOST Community Edition v0.1.0

首个社区版发布：本地优先、Windows-first 的 Live2D 数字人插件栈，面向 DeepSeek Harness（DSH），以 Apache-2.0 三组件源码快照发布。

First release of the community edition: a local-first, Windows-first Live2D digital-human stack for DeepSeek Harness, published as a three-component source snapshot under Apache-2.0.

Highlights

• Real-playback lip sync — the mouth follows a ~20 ms RMS envelope of the PCM actually handed to the audio device, with measured output-latency alignment.
• Independent DSH bundles — the Live2D pet and the TTS bridge can be enabled or disabled separately; disabling the bridge restores native TTS.
• Local-only — no cloud TTS, no accounts, no telemetry; services bind to loopback only.
• One reference voice per install — voicesDir holds one reference clip plus its same-stem transcript; emotion names are event labels, not separate voices.
• Windows-first installer — install.ps1 renders placeholders and verifies module resolution; uninstall.ps1 is reversible.
• Docs + tests — architecture/install/licensing docs, offline CI, Python TTS tests.

Components

Directory	Purpose
live2d-pet/	Live2D desktop pet: DSH bundle + plugin + transparent Electron window
tts-bridge/	DSH bundle + Python shim that tees the real playback PCM envelope/emotion to the pet
voice-plugin/	Snapshot of @deepseek-ai/dsh-tool-tts 0.1.5-rc.2 with the VoxCPM 1.5 Python engine (MIT)

Notes

• Windows 10/11 tested; Linux/macOS untested.
• No models, Cubism Core, VoxCPM weights, reference audio or node_modules are shipped.
• Personal use is free; commercial editions/features require author authorization (see README).