Writing with Your Voice: A Dictation Mode Workflow Guide
Press a shortcut, talk, and Audio Note types into the field you are using; voice commands let AI rewrite a selection. Setup, workflow, and when typing wins.
Why talk instead of type
Many people get ideas out more fluidly by speaking, while typing invites interruptions from wording, formatting, and the cursor. Dictation is not a shortcut; it gets the first draft out before your editor brain takes over.
Audio Note dictation transcribes speech on your computer; audio is not sent to the cloud.
Status: beta in v2.2.0, on every plan
Dictation ships as a beta starting with v2.2.0 and is available on Free, Standard, and Pro. Beta means it is open to everyone while details keep changing based on feedback.
The plans differ in three places:
| Free | Paid plans | |
|---|---|---|
| Dictation and voice commands | Yes | Yes |
| Whisper models for dictation | Tiny and Base family | Models included in your plan |
| Realtime dictation | No | Yes |
| Save dictation records to a workspace | No | Optional |
See the pricing page for the full plan breakdown; setup steps, the plan comparison table and FAQs live in the dictation docs.
What dictation does
Dictate. Click into a text field, press the dictation shortcut, and talk. Release or press again to finish. The transcript is inserted at the caret: email, chat, notes.
Voice command. Select some text, press the voice-command key, and say what you want, such as "make it more formal", "cut it to three sentences", or "translate to English". AI replaces the selection with the result. Questions get an answer card instead of a replacement.
Setup
- Turn on dictation in settings and pick a model and language. Download a model if needed.
- Set the shortcut: a preset or your own combination, triggered by short press, long press, or double press. The app warns about conflicts with system or app shortcuts.
- Choose how text is typed: character by character, like fast typing, or instantly as a whole phrase.
- Decide on AI polish. With it on, the transcript goes through AI before insertion; with it off, the raw transcript is inserted.
- On macOS, grant Accessibility under System Settings → Privacy & Security so text can be typed into other apps.
"Realtime dictation" shows text in the dictation window while you speak; off means text appears after you finish. It is not available on the Free plan.
Besides character-by-character and instant typing, there is an "Auto" option: short text is typed character by character, long text is inserted at once.
Which model to use
Dictation is latency-sensitive. Waiting several seconds after each sentence ruins the flow, so smaller and faster usually wins.
If you have run the Performance Benchmark from the home screen, the dictation model picker marks the model that suits your computer as recommended and flags Whisper models that are heavier than your machine handles well. How the benchmark picks models is covered in From Zero to AI Notes.
- On Free, pick Tiny or Base. Fine for everyday sentences; double-check names and jargon.
- For mostly Chinese, try a Chinese realtime model or a Chinese-friendly model such as SenseVoice.
- Each model downloads once and then works offline.
What it is good for
First drafts. Say the ideas; the transcript becomes the skeleton.
Notes and outlines. Speaking pushes loose thoughts into full sentences.
Brainstorming. Talking through a block often breaks it; pick the useful ideas afterwards.
Email and messages. With AI polish or voice commands, casual speech becomes clean writing.
A talk-first workflow
- Say the whole piece without stopping to edit.
- Skim the transcript and mark what to keep, move, or cut.
- Restructure and fill gaps.
- Select passages and use voice commands to adjust tone or length.
- Do a final pass yourself.
AI templates for dictation
| Template | What it does |
|---|---|
| Simple dictation | No extra prompt; inserts the transcript |
| AI polish | Re-segments sentences, fixes typos, improves structure |
| Punctuation | Adds punctuation and fixes homophone errors without changing the wording |
| AI email | Formats as an email with greeting and closing |
| AI Q&A | Treats what you said as a question and answers it |
| AI translate | Translates to English, or to Chinese if the source is English |
AI polish uses your configured backend. Local backends (Ollama, LM Studio) keep text on your computer; cloud APIs send it to that service.
Tips
Focus the field first. Click into the target field before the shortcut, and do not switch windows during character-by-character typing.
Quiet room. A laptop mic works in a pinch; a headset or external mic is steadier for long sessions.
Read back when stuck. Rereading the last few sentences keeps momentum.
First words may lag. With on-demand loading, the model loads at the start of each session.
When to dictate
| Writing | Dictate? | Why |
|---|---|---|
| Drafts, essays, scripts | Yes | Flow first, edit later |
| Meeting recaps, journals | Yes | The spoken voice is the point |
| Email and messages | Yes | Works well with AI polish |
| Technical docs | Partly | Speak the concepts; type code and formatting |
| Citation-heavy writing | Not really | Needs exact quotes and wording |
| Tables, formulas, layout | No | Structured content is easier by hand |
FAQ
Nothing happens when I press the shortcut. Make sure a text field is focused, dictation is on, and a model is downloaded. On macOS, check Accessibility.
I finished but no text appeared. Focus may have moved during typing. Click back into the field and retry; the dictation history keeps the record.
Which plans include dictation? All of them, including Free. On Free, dictation models are limited to Whisper Tiny and Base, realtime dictation is off, and dictation records are not saved to a workspace.
Is dictation final? It ships as a beta in v2.2.0. It is available now; details will keep improving.
Can I skip AI? Yes. With AI off, the transcript is inserted directly. Voice commands need AI.
