@toddzheng024/vapor-voice-local
Vapor 的可选 macOS 本地参考音色准备
插件会安装到这里；不确定时保持 web。
npx -y @deepseek-ai/dsh plugin --profile web add @toddzheng024/vapor-voice-local@0.3.0Vapor Local Voice 以 @toddzheng024/vapor-voice-local 发布，当前版本为 0.3.0。Plugin Hub 会校验它的 manifest，并保存精确安装来源，便于复现安装结果。
Prepare a 3–30 second reference from a registered audio/video revision. This package is separate from the cross-platform Studio Profile because its network-denied execution relies on macOS sandbox-exec.
Install the requirements into your own Python environment and set VAPOR_VOICE_PYTHON to its executable. FFmpeg is required. Prepare Demucs weights separately before use; processing denies network and will not download missing weights. The package contains processing code, never recordings, models, global voice profiles or credentials.
After processing, listen to all tracks, verify the transcript and record the source authorization in Lab. Saving the legacy reference Profile does not synthesize speech or establish voice quality.
The 0.3 tool catalog separately exposes voice.add, voice.list, voice.doctor, voice.speak and the user-approved voice.prepare-model download. These use the existing native profile format under the selected project's voices/ and locked weights under models/; they do not consume the legacy Lab reference-asset JSON. Choose VAPOR_TOOLS_PYTHON (or plugin config.python) with MLX-Audio installed for these operations. Model-dependent inference has not been accepted by full listening. See operations.