speaklisted

# speak — Text-to-Speech via speak.exe Speaks text out loud using a single self-contained native binary. No Python, no `ffplay` pipeline. A glowing orb appears in the bottom-right corner of the screen while the audio plays, optionally captioned with what the utterance is about. Source: `C:\Dev\www\claude-speak` (published at https://github.com/Dovyski/speaker). This skill lives in that repo under `skill/speak/`; every path below assumes the checkout is at `C:\Dev\www\claude-speak` — adjust them if it is cloned elsewhere. The repo's own docs go deeper than this skill needs to: `docs/references/speech.md` (engine, voices, daemon), `docs/references/overlay.md` (orb and captions), `docs/references/pointing.md` (rings and targets) and `docs/references/attention-panel.md` (the per-terminal card `--session` resolves against). ## Speaking ```bash "C:/Dev/www/claude-speak/speak.exe" "The text to speak." ``` That is the whole thing: it synthesizes, plays through the default audio device, and animates the orb. The call is synchronous — it returns when playback finishes, so do not background it if you want to know it completed. The user can **click the orb to pause** playback and click it again to resume, so a call may take longer than the audio lasts (it prints `speak: paused` / `speak: resumed` on stderr). A paused utterance keeps the process alive until it is resumed, so a call that is killed by a command timeout may simply have been held — not a failure to investigate. Model