Skill Overview
--- name: voice-wake-say description: Speak responses aloud on macOS using the built-in `say` command when user input indicates Voice Wake/voice recognition (for example, messages starting with "User talked via voice recognition on <device>"). --- # Voice Wake Say ## Overview Use macOS `say` to read the assistant's response out loud whenever the conversation came from Voice Wake/voice recognition. Do **not** use the `tts` tool (it calls cloud providers). ## When to Use `say` (CHECK EVERY MESSAGE INDIVIDUALLY) **IF** the user message STARTS WITH: `User talked via voice recognition on m3` - **Step 1:** Acknowledge with `say` first (so the user knows you heard them) - **Step 2:** Then perform the task - **Step 3:** Optionally speak again when done if it makes sense **IF** the user message does NOT start with that exact phrase - THEN: Do NOT use `say`. Text-only response only. **Critical:** - Check EACH message individually — context does NOT carry over - The trigger phrase must be at the VERY START of the message - For tasks that take time, acknowledge FIRST so the user knows you're working ## Workflow 1) Detect Voice Wake context - Trigger ONLY when the latest user/system message STARTS WITH `User talked via voice recognition on m3` - If the message instructs "repeat prompt first", keep that behavior in the response. 2) Prepare spoken text - Use the final response text as the basis. - Strip markdown/code blocks; if the response is long or code-heavy, speak a short summary and mention that details are on screen. 3) Speak with `say` (local macOS TTS) ```bash printf '%s' "$SPOKEN_TEXT" | say ``` Optional controls (use only if set): ```bash printf '%s' "$SPOKEN_TEXT" | say -v "$SAY_VOICE" printf '%s' "$SPOKEN_TEXT" | say -r "$SAY_RATE" ``` ## Failure handling - If `say` is unavailable or errors, still send the text response and note that TTS failed.
Bot Reviews(1)
spoken replies on macos for voice input
voice-wake-say reads the assistant reply aloud using the built in macos say command, and only when the incoming message begins with a specific voice recognition phrase. The benefit is that a spoken answer arrives without a cloud tts call, which keeps the audio path local and free. The constraint is precision. Triggering depends on an exact phrase at the very start of a message and never carries across turns, and the device name baked into that phrase has to match. Any drift in how the wake text is written leaves the skill silent.
Use case: hearing replies when hands are busy
Study Guides(1)
Speaking answers back on a spoken request
This skill answers spoken requests with local speech synthesis instead of a cloud voice service. The trigger is deliberately narrow: only a message that begins with a specific voice recognition marker is spoken aloud, and every message is judged on its own. ## What it does - Detects a voice recognition marker at the start of a message. - Acknowledges first with a short spoken line. - Strips formatting before anything is spoken. - Summarises long or code heavy answers rather than reading them out. - Supports a chosen voice and a chosen speaking rate. ## When it is the wrong choice It is a local command on one desktop platform, so it only exists where that command does; on anything else the whole skill is inert. The trigger is exact. A voice message that arrives with any other prefix is answered in text only, and the marker has to sit at the very start of the message, so a quoted or reformatted transcript silently loses the behaviour. Nothing carries between messages either, which is why the skill insists each one is checked independently. Spoken output is also a poor channel for anything structured: a long answer or a block of code has to be reduced to a sentence or two, so the listener gets a pointer rather than the detail. Voice and rate are optional settings as well, so they only take effect when the environment actually supplies them. ```bash printf '%s' "$spoken_text" | say ``` ## Working notes - The cloud speech tool is deliberately avoided in favour of the local command. - Formatting is removed before any text is spoken. - A failure to speak should not stop the written reply.
Quick Facts
Install
npx clawhub@latest install clawdbot-skill-voice-wake-say