In short
Live transcription turns speech into text as it is spoken, so you can follow along, search the meeting while it is still going, and see a transcript the moment it ends.
Transcribing afterwards works from the finished recording, which lets a recogniser use the whole context and take its time, and usually gives a more accurate result. You only get the second if the recording was kept.
This guide explains the trade-off, why the two differ, and what having both needs.
The short version
| Live | After the meeting | |
|---|---|---|
| When you see text | As people speak | Once the pass has finished |
| Context the recogniser can use | What has been said so far, and a little after | The whole recording, before and after each word |
| Time it can spend | Must keep up with speech | As long as it needs |
| Needs the audio kept | No | Yes |
| Good for | Following along, catching a point you missed, notes straight away | The version you keep, search later, and quote |
Why live transcription is harder
A live recogniser has two constraints that a recogniser working on a file does not.
It cannot wait for the end of the sentence. Many words are only clear from what comes after them. "Their", "there" and "they're" sound the same; "fifteen" and "fifty" can differ by one syllable that is easier to judge once the rest of the sentence is known. A live recogniser has only a short look ahead, so it guesses and then revises.
It has to keep up. Whatever it does per second of audio has to take less than a second, on whatever else the computer is doing during a call. That limits how large a model it can run and how many alternatives it can weigh.
To make the wait feel shorter, many live systems show a first guess immediately and correct it as more audio arrives. Apple's Speech framework, for example, can report "volatile" results, described in its documentation as immediate, rougher guesses, alongside finalised results that will not change (Apple Developer: SpeechTranscriber.ReportingOption.volatileResults, checked 25 September 2026). That is why words in a live transcript can shift as you watch.
Why working from the recording helps
Given a finished recording, a recogniser can look at the words both before and after any point, use a larger model because nobody is waiting, and go back over hard stretches. It is also not competing with the call app for the computer's attention. Published accuracy figures are usually measured this way, on complete files, which is one reason they can look better than what you see live; word error rate, explained covers how those figures are made.
After the meeting is also when other work that needs the whole recording can happen, such as telling several voices on the other side apart.
None of this makes a post-meeting transcript perfect. The hard words, names, numbers and jargon, stay hard; see why meeting transcripts get names and numbers wrong. The audio itself sets a ceiling that no second pass can lift.
You need the recording to have both
This is the practical point. A second pass needs the audio. Notetakers make different choices here:
- Some stream audio to a transcription service during the meeting and do not keep it. You get a live transcript, and that is the transcript. The comparison of Notey and Granola describes one tool that works this way, from its own documentation.
- Some record the meeting and transcribe the file afterwards. You wait for the transcript, but it can be the more accurate kind.
- Some do both: transcribe live, keep the recording, and improve the transcript after the meeting.
Keeping the recording has a cost as well as a benefit. It is a file of everyone's voices that you are responsible for: where it is stored, who can reach it, and when it is deleted. Discarding audio is a legitimate choice for people who would rather have no audio files at all. The question to ask of any tool is which it does, and where the audio goes in each case.
What changes when you use each
During the meeting, a live transcript is for orientation: glancing back at what was said a minute ago, checking a name as it was spoken, or noticing that a decision has just been made. Treat it as a draft that is still settling.
After the meeting, the transcript is what you quote, search in three months and hand to an AI to summarise. That is where accuracy pays off: a summary inherits every error in the transcript it was written from.
How Notey does it
Notey does both, on the Mac. On-device transcription explains what that means for privacy and accuracy.
- Live: your microphone and what the Mac plays are transcribed as the meeting happens by Apple's speech recognition. Words appear as they are spoken, and a phrase the recogniser is still revising is shown lighter until it settles.
- Kept: both tracks are recorded to your disk. Click a line's timestamp to hear that moment.
- After an English meeting: Notey reads the recording again on your Mac and keeps the words the recognisers agree on. It is English only; meetings in the other languages keep their live transcript. Why Notey re-transcribes a meeting after it ends covers when it runs, what it downloads, and how to switch it off.
Frequently asked questions
Is live transcription less accurate than transcribing a recording?
Usually somewhat, with the same kind of model. A live recogniser has to commit to words with little of what comes next, and has to keep up with the speaker. Working from a finished recording, it can use context from both sides and take as long as it needs.
Why do words in a live transcript change as I watch?
The recogniser shows its first guess straight away and revises it as more audio arrives. Apple's speech framework calls these volatile results; they are replaced by finalised text once the recogniser is confident.
Can I get a better transcript after the meeting if my notetaker did not keep the audio?
No. A second pass needs the recording. If a tool transcribes live and discards the audio, the live transcript is the only one there will be.
Does Notey transcribe live or after the meeting?
Both. It transcribes live on the Mac with Apple's speech recognition, and after an English meeting it reads the recording again on your Mac and keeps the words the recognisers agree on.