Transcripts and speakers
How Metis AI transcribes on your device, which languages it understands, and how to name and correct who said what.
Updated 13 November 2026
Metis AI turns a recording into a transcript split by speaker. It happens on the device that made the recording, so your audio isn’t sent anywhere to be transcribed. This article covers languages, names and how to fix a line that came out wrong.
How transcription works
While you record, a live transcript appears as people talk. When you stop, Metis AI listens to the whole recording again and writes a more accurate transcript, then works out who said each line.
The speech model downloads once, the first time you need it (on Wi-Fi when it can). After that, transcription works offline.
Metis AI picks the speech model for your device. iPhone 15 and later, on iOS 18 or later, use the newer Qwen3-ASR model once a one-time check shows it runs well; other devices use Whisper Large. There’s nothing to set.
The phone transcribes with Qwen3-ASR, the same speech model as the iPhone (about 0.9 GB). You can see it inSettings, under On this phone › Transcription.
Open Account and go to Recording on this computer. Transcribe with Qwen3-ASR is on by default. Under it, choose the Whisper model to use if Qwen3-ASR is off or can’t run on your computer. A meeting transcribed with Whisper says so.
The website doesn’t record or transcribe. It shows the transcripts your phone or computer synced to your account.
Choose the meeting’s language
Open Settings and choose a Language. It applies to both the live transcript and the final one.
- Auto-detect lets Metis AI decide.
- 中文 + English is for meetings that switch between Chinese and English. The live transcript shows Chinese, and the final transcript detects each sentence’s language.
- English, 日本語 (Japanese) and 한국어 (Korean).
- 中文 (简体) or 中文 (繁體) for Chinese in simplified or traditional characters.
There’s no language setting. The phone detects the language of each meeting by itself.
In Account › Recording on this computer, set Language to Detect automatically, English or 中文.
The notes are written in the language of the meeting. The apps themselves are in English.
Teach it names and jargon
If Metis AI keeps misspelling a person, a company or a product, add it as a custom word. When something in a transcript sounds like one of your words, it’s written your way.
- On your iPhone or iPad, open Settings and tap Custom words.
- Add each name or term the way it should be spelled.
Custom words stay on that device. They’re only on iPhone and iPad for now; the Android and Windows apps don’t have them.
How speakers are told apart
Speaker separation runs on your phone or computer. Each voice gets a label — Speaker 1, Speaker 2 and so on — until you name it. On Windows, Metis AI also tells you (the microphone) apart from the people on the call.
On iPhone and iPad, if it found too many or too few people, open the meeting, tap ⋯ and use the Speakers item to choose Auto-detect or the number of people, from 2 to 8. Metis AI then sorts the voices again. To join two labels that are really one person, use Merge speakers in the same menu.
Name the speakers
After a meeting is processed, Who was talking? opens. Play a few seconds of each voice and type a name, or pick someone you’ve named before. Tap Done, or Skip for now.
To do it later, tap ⋯ › Name speakers, or double-tap a name in the transcript.
Open the meeting’s Transcript tab, click a speaker above the transcript, type a name or choose one under People you’ve named, and click Save.
Open the meeting’s Transcript tab, click a speaker above the transcript, type a name or choose one under People you’ve named, and click Save.
You can’t name speakers on Android yet. Names you give on your iPhone, iPad, computer or the web appear on the phone once the meeting has synced.
Naming people matters for your tasks: each task’s owner is a speaker, so a name means the task lands on the right person.
Remember a voice for next time
On iPhone and iPad, Metis AI can remember the voice of someone you’ve named and name them by itself in later meetings. It’s off until you turn it on for a person.
- In Who was talking?, name the person and tap the waveform button beside the name.
- Metis AI asks whether that person agreed to have their voice remembered. Tap They agreed only if they did.
In later meetings, a speaker it recognised shows Identified by voice. If it got someone wrong, tap Not …? beside the name and Metis AI learns from it.
To see or remove voices, open Settings › Remembered voices. Tap a voice to play its samples, or swipe to forget it. A voice that isn’t used for 12 months is forgotten automatically.
Fix a word or who said a line
On iPhone and iPad, open the meeting’s transcript and touch and hold a line. You can:
- Edit text to fix a transcription mistake. Only that line changes.
- Change speaker to move the line to someone else, or to a New speaker. A banner lets you undo it, and may offer Use it to help recognise that person’s voice next time.
- Create task from this when a commitment was missed.
- Copy the line.
Tap a speaker at the top of the transcript to show only their lines. After moving lines, you may want to Regenerate summary & tasks from the ⋯ menu so the notes follow.
Editing lines isn’t available on Android, Windows or the web yet. Edits you make on your iPhone or iPad sync to your other devices.
Hide a line before you share
Some things said in a meeting shouldn’t go any further. On iPhone and iPad, touch and hold the line and tap Hide from sharing. To undo it, choose Show when sharing.
A hidden line stays in your own copy, but it’s left out of exported files, chat messages, meeting links, the website’s transcript and anything shared to improve Metis AI. When you export a file you can also tap Remove parts before sending and tap lines to hide them.
Transcribe a meeting again
If a transcript came out badly — for example, the wrong language was set — you can have Metis AI listen to the recording again. This needs the recording on that iPhone or iPad.
- Open the meeting and tap ⋯ › Transcribe again.
- Confirm with Transcribe again.
You get a new transcript, speakers and notes. Names you gave carry over where Metis AI finds the same person, tasks you kept, changed or set aside stay, and hidden lines stay hidden. Edits to the transcript’s text are replaced.
The option isn’t in the menu
It only appears when the recording is on the device. If you used Delete audio, keep notes, or the meeting was recorded on another device, there’s nothing to listen to again.