Speech preferences and vocabulary

Open Settings → Speech & Transcription to adjust available voice behavior. Hosted speech services are managed; preferences shape how you interact with them.

Speaking and listening

Use available text-to-speech settings for voice persona and speaking speed. Some providers have their own preset voices and fixed pace, so controls differ. A missing speed control does not mean speech is unavailable.

Hands-free dictation

Set confirmation phrases that end a request and send it, and exit phrases that discard accumulated dictation. Test them with harmless text before depending on them. The ordinary mic button remains push-to-talk.

Mute while speaking prevents Gideon’s own reply from being captured. The echo filter provides another check against speaker bleed. Keep these enabled when using speakers unless you have a reason to change them.

Clean spoken output

The clean-text option can strip code blocks and simplify long URLs and paths for speech while retaining the full written transcript. Voice-origin framing tells the model that an input was dictated so it can account for transcription mistakes.

Vocabulary and corrections

Add specialist terms or names to the personal vocabulary. Review automatically derived terms and prune incorrect ones. Resync from the knowledge graph when relevant names change. Corrections learned from transcript edits can be marked always when you want the substitution applied again.

Shortcuts and platforms

A global push-to-talk shortcut requires the desktop capability; browser controls are limited to the browser. Check microphone permissions and the actual input device before changing model-related settings.