Voice
Dictate review comments, have findings and answers read back to you, and review a pull request hands-free in the Ultra Zen console. Pro.
DiffGuardian can listen and it can talk back. Voice is useful for the part of reviewing that is not typing — dictating a comment while you are still looking at the code, or having a long walkthrough read to you.
Available on Pro. Both directions are off by default and are switched on separately in Settings → Voice.
#Input — speech to text
Turn on Input voice and a Dictate button appears in the inline comment composer. Speak, stop, and what you said is written up into a clean comment — see Writing review comments. Voice input also drives the Ultra Zen console below.
Two settings shape it:
- Language — English (US or UK), Spanish, French, German or Japanese.
- Auto-send after pause — send what you said when you stop talking, instead of waiting for you to press send.
Speech recognition uses your system's speech services, so it needs microphone permission. On macOS, grant it under System Settings → Privacy & Security → Microphone; if dictation never starts, that permission is the first thing to check.
#Output — text to speech
Turn on Output voice and DiffGuardian will read things aloud: the findings in Replay mode, which walks the review one finding at a time and highlights the hunk each one is about, and the answers in Ultra Zen. Two providers:
| Provider | Notes |
|---|---|
| System | The voices your operating system already ships. No key, no network, no cost. |
| OpenAI | Higher-quality voices, using your own OpenAI API key. Choose a voice and a TTS model. |
The OpenAI option reuses the key you already entered in Settings → AI providers; there is no separate key to manage. Like every other cloud call, it goes from your machine directly to the provider.
#Ultra Zen — the voice console
With voice input and output on, a pull request can be reviewed hands-free in the Ultra Zen voice console. Hold the space bar to ask a question, release to send, and the answer is spoken back. It will summarise the features in the change, walk you through one of them, or answer an ad-hoc question about the pull request.
If Ultra Zen tells you it needs voice, one of the two toggles is still off.
#Where the audio goes
Speech recognition happens through your operating system. Spoken output is either synthesised locally by the system voice, or generated by OpenAI with your own key. Neither direction touches DiffGuardian's servers. See Privacy and data handling.