Purpose
Read a conclusion/summary aloud in natural Taiwan Mandarin speech, achieving the conversation loop of "user voice input → Claude voice reply".
- Generate + play:default streaming(Edge-TTS generates and plays as it goes, first sound in about 1 second; ffplay/mpv pipe playback, no window)
- Fallback chain: streaming → full file (MediaPlayer/WMPlayer COM) → SAPI offline; never
Start-Processopen a player - Specify a particular person's voice (e.g. "use ○○'s voice")and the environment already has a voice cloning skill installed (e.g.
voice-clone)only then switch to that skill; if not, still use this skill
Execution steps
1. Write the script
- Colloquial, concise (100–250 characters), numbers in Chinese ("fifty pages" not "50 pages")
- Only say the conclusion and next step; leave details in the text reply
2. Generate + play (one command, ⚡ by default run in the background)
pwsh -NoProfile "<本技能資料夾>\speak.ps1" "<講稿文字>"
（<本技能資料夾>= the directory where this SKILL.md is located, i.e. the base directory shown when the skill is loaded)
Background parallel (important): the user wants "the voice to pick up right when the text reply is finished"—so when calling,if the platform supports background execution (Claude Code's Bash run_in_background: true, equivalent approaches in other agents), throw it into the background, and immediately start writing the text reply; the voice will play while you type. When the background completion notification comes back, take the first_sound_s／total_s from the output and report it along the way.
Other options:
- For long scripts change to
-File "<講稿.txt>" - default voice
zh-TW-YunJheNeural(male); female voice-Voice zh-TW-HsiaoChenNeural - User says "save it" → add
-Out "<專案路徑>.mp3"(use full-file mode and keep the audio file) - Streaming does not produce a file; only use
-Out
3. Report
One sentence is enough (mode, first-sound latency, audio file location if saved). Do not repeat the script content.
Notes
- If the user's sentence contains "use ○○'s voice" → if a voice cloning skill is already installed (e.g.
voice-clone) use it instead, otherwise still use this skill and explain - Edge-TTS requires the internet; when offline it automatically falls back to Windows SAPI (more robotic but zero latency)
- Do not use
Start-Processto play—it opens an external program window, which the user explicitly does not want