Voice: talking to Pinchy and hearing him answer

You can talk to Pinchy instead of typing, and he can answer out loud. Speech is recognised on your computer; nothing you say is sent anywhere.

What to ask for

Hey Pinchy: say his name and your question in one breath (“Hey Pinchy, what time is it”), hands free. He hears you out and it is sent when you stop talking; say only his name and nothing happens. If you would rather read your words before they are sent, choose “Your words go into the chat box” in the settings. A mic light shows while he listens. He never listens while the screen is locked, and not to himself while he is speaking. You can change the words he listens for.
Push to talk: hold a key, speak, let go. A quick tap keeps listening until you pause. What you say can go to Pinchy as a question, or be typed for you into the app you are in. With Hey Pinchy on, the microphone is already open, so nothing is cut off at the start.
Dictation: the microphone button in the chat turns speech into text you can edit before sending.
The Humotranslator (his voice): Pinchy is a lobster, so he clicks — the Humotranslator, the little box he holds up while he answers, says it out loud in a human voice, either only when you talked to him or always. Its meter and sound waves move with what is being said. Long answers are shortened to the first sentences and code is never read out.

A small speech model runs on this computer. It is downloaded once (about 57 MB) the first time you switch a voice feature on; the recording is deleted right after it is turned into text.
The Humotranslator’s voice also runs on this computer. On Linux, and for the nicer voice on Windows, it is a one-time download of about 90 MB; macOS uses its own voice.

Settings → Voice & writing → Hey Pinchy: switch it on, the words he listens for, what happens after you say his name (he answers, or your words go into the chat box), which model listens, test the wake word, remove the model.
Settings → Voice & writing → Push to talk: switch it on, the key to hold, ask Pinchy or type it for me, quick tap to keep listening, the longest take.
Settings → Voice & writing → Dictation: speech model (Base, Small, Best accuracy), language, show words while I talk.
Settings → Voice & writing → Humotranslator: answer out loud (Off, When I talk to him, Always), the voice, style, speed, keep listening after he answers.
Settings → Pinchy → Shortcuts: every shortcut in one place, including push to talk.
Settings → Safety & data → Check everything: tests the microphone and the shortcuts and offers a fix.

Hey Pinchy and push to talk are off until you switch them on.
In a loud room (a video, music, a call) he checks what he hears a little less often; he does not stop listening. If he does not react, “Check everything” says why, for example that the microphone gives no sound.
He listens for his name in the language you dictate in. If you speak several languages, set the dictation language to the one you call him in.
The Base model is the fastest; Best accuracy is slower and bigger. Pick the language if he mishears a language other than English.
If a shortcut does nothing, another app may hold the same keys: change it under Shortcuts, or run “Check everything”.