| 1 | --- |
| 2 | name: tts |
| 3 | description: Speak or render text to speech locally. Use when: read aloud, speak, voice output, tts, text to speech, or an audio version of text. |
| 4 | invocation: model+user |
| 5 | --- |
| 6 | |
| 7 | # Text to Speech |
| 8 | |
| 9 | ## When to use |
| 10 | Reading text aloud or rendering speech audio files, using only the voices |
| 11 | already on the machine. |
| 12 | |
| 13 | ## Setup |
| 14 | No account or API key. Pick the platform command; fail loud when the |
| 15 | Linux engine is missing (`sudo apt install speech-dispatcher` or `espeak-ng`): |
| 16 | |
| 17 | - macOS: `say -v Samantha -o out.aiff "text"` (list voices: `say -v '?'`) |
| 18 | - Linux: `spd-say "text"` or `espeak-ng -w out.wav "text"` |
| 19 | - Windows: PowerShell `System.Speech.Synthesis.SpeechSynthesizer` |
| 20 | |
| 21 | ## Workflow |
| 22 | 1. Keep utterances short; split long text at paragraph breaks. |
| 23 | 2. Confirm voice and speed when the user cares; otherwise default voice, normal rate. |
| 24 | 3. For files, report the path, format, and duration. |
| 25 | |
| 26 | ## Non-goals |
| 27 | - Do not clone voices or impersonate a real person. |
| 28 | - Do not call cloud TTS APIs (different cost and privacy posture — ask first). |
| 29 | - Do not play audio unprompted in shared environments; render a file instead. |
| 30 |