# Voice and meetings

> Dictation, reading aloud and live transcription on your Mac, and recording a meeting into a note with a summary, decisions and action items.

Applies to Canto 1.0 beta 3 · Updated Oct 10, 2026 · https://lonelyduck.io/docs/canto/voice

Taken from the app's source for Canto 1.0 beta 3: `src/shared/whisperCatalog.ts`.

Everything on this page happens on your Mac. Dictation and live transcription use Whisper models that run in Canto; reading aloud uses the voices that come with macOS; meetings use macOS's own speech recognition and your model on the Mac. No audio is ever uploaded.

## Turn voice on

Dictation and reading aloud are off until you turn them on: open **Settings › Models › Voice** and switch on **Dictation**. Then get a dictation model from the list below it (**Get** downloads it from Hugging Face; the download says how big it is and can be cancelled):

| Model                                 | Languages    | In Canto's words                                                                         |
| ------------------------------------- | ------------ | ---------------------------------------------------------------------------------------- |
| Whisper Large v3 Turbo                | 99           | Fast and accurate; the default                                                           |
| Whisper Large v3 Turbo · Korean-tuned | Korean only  | Listed first when you dictate in Korean; about 16% errors on its test set, down from 24% |
| Whisper Large v3                      | 99           | The most accurate, and the slowest                                                       |
| Whisper Medium                        | 99           | Smaller and a little less accurate than Large v3 Turbo                                   |
| Whisper Large v3 · Italian-tuned      | Italian only | Fine-tuned on Italian speech                                                             |

"99 languages" is OpenAI's own claim; it's strongest in about ten of them.

The same row has **Dictation language** and **Try it**, which listens for five seconds and shows what the model heard. Its **⋯** menu has **Add a period at the end**.

The first time Canto records, macOS asks for the microphone. If you said no, the **Microphone** row in the same section says so; turn it on in **System Settings › Privacy & Security › Microphone**.

## Dictation

Press `⇧⌘V` where you're typing to start, and again to stop. In a note, the words go into the note. In Ask, press the microphone, which is the composer's button while the message field is empty. ⌘K's **Toggle dictation** works too.

You can change the keys in **Settings › Keyboard**.

## Read aloud

Select text and press `⇧⌘R` to hear it, or press it with nothing selected to hear the whole note. ⌘K has **Read selection aloud**, **Read this document aloud** and **Stop reading aloud**.

Canto uses macOS's voices. Choose one, and its speed and pitch, under **Read aloud** in **Settings › Models › Voice**.

In Ask's message field, `⇧⌘R` turns Think on or off instead; see [Keyboard shortcuts](https://lonelyduck.io/docs/canto/shortcuts#same-keys-different-places).

## Live transcription

`⌘T`, or **/ › Live transcription…** in a note, opens a panel that turns speech into text as you talk:

* **Start recording**, then **Stop** when you're done. Text appears in chunks of about 10 to 30 seconds, using your dictation model.
* **Copy** copies the transcript.
* **Insert into note** puts it in the note; while recording, **Stream into note** adds it as it comes.

It needs a dictation model first; the panel says so and links to **Settings › Models › Voice**.

## Meetings

**Meeting capture** records a meeting on your Mac and writes a note from it. It's part of **Pro**, and each Mac gets three free tries.

### Record a meeting

Start one with ⌘K's **Start meeting**, or right-click **Meetings** in the sidebar's Library and choose **Record a meeting**. Before anything is recorded, a sheet shows:

* that the meeting stays on your Mac, and a reminder to tell everyone you're recording (in some places you need their consent),
* what to record: **My microphone**, or **Microphone and this Mac's sound** (the other people on a call),
* the **Language**: the Mac's language, English, or 한국어 (Korean),
* what happens to **The recording**: **Keep it for 30 days**, **Keep it**, or **Delete it once it's transcribed**,
* and anything that would stop it working, such as a permission macOS refused.

**Start recording** begins. If macOS needs its speech model for that language, the key says **Download and record**: macOS downloads it once, from Apple. Your audio never leaves your Mac.

### While it records

A capsule with a crimson dot and the time stays on screen above everything, so you never forget the microphone is on. It has a level meter for **You** (your microphone) and **Others** (the Mac's sound), **Pause** and **Resume**, **Stop**, and **Cancel**.

After **Stop**, the capsule says what Canto is doing (transcribing, **Writing summary…**, **Saving the note…**) until the note opens. If Canto quits during a meeting, it offers to finish it the next time it opens.

### The meeting note

The note has a **Summary**, **Decisions**, **Action items** and the **Transcript**, plus the recording if you kept it. The summary is written by your model on the Mac. Meeting notes get the `meeting` tag, and **Meetings** in the sidebar's Library lists them, newest first.

### What meetings need

* **Transcripts need macOS 26 or later.** On macOS 14 and 15 the recording is attached to the note without a transcript.
* **Recording the Mac's sound needs macOS 14.4 or later**, and macOS asks you to allow it. Without it, only your microphone is recorded.
* Transcripts don't tell the other people on a call apart: everyone on the Mac's side is "Others". There's no live word-by-word text while you record.

Meeting defaults are in **Settings › Memory & privacy › Meetings**: **Keep recordings** (30 days), **Record this Mac's sound** (on) and **Meeting language** (Automatic). Meetings are never sent to a server or a cloud model.