Transcripts and speakers

Transcription

Settings → Summaries chooses the model that turns audio into text, and whether new audio is transcribed automatically.

Models other than Apple's download the first time you use them (the list shows what's installed). Only an installed model can be the default.

These models run on your Mac, and your audio isn't sent anywhere. You can also choose Fireflies.ai, Pocket or Plaud as the transcription engine (Settings → Summaries); then the audio goes to that service, which makes the transcript. See Which should I use?.

Transcribe again

On a meeting, Transcription (⇧⌘T) transcribes it if it hasn't been, or, if it has, lets you choose a model and Retranscribe. Use this if the first result was poor, or after you switch to a better model.

Where the transcript shows

The Transcript tab lists each line with its time and speaker; a note under the tabs says which model or service made it ("Transcribed by …").

While a service is making the transcript, the tab says Being transcribed by Pocket (or Fireflies.ai, or Plaud) with a moving bar, instead of offering to transcribe it here. A strip under the meeting's title says what the service is writing (for example "Plaud is writing the transcript and summary") and its latest step. The wait doesn't hold anything else up: you can import, record and transcribe other meetings meanwhile. The transcript appears when the service is done. If the service can't do it, the meeting is transcribed on your Mac instead, and the tab shows that progress.

Speakers

While transcribing, MacParsley separates the voices and labels them Speaker 1, Speaker 2… It then compares each voice with the voices it has learned ("voice prints") and names the ones it recognizes.

Transcripts from Plaud, Pocket or Fireflies

When one of these services makes the transcript (an imported recording, or a service chosen as the transcription engine), two different jobs happen, in two different places:

Your meeting isn't transcribed twice. Your Mac doesn't write down any words for these meetings; it only listens for who is talking, which is a much smaller job (a minute or two for an hour of audio, instead of a full transcription). You may see Separating speakers in the activity area, or notice your Mac working while the service is busy: that's this step.

Why do it on your Mac, when the service already labels speakers? Because naming people depends on it. To recognize someone, MacParsley compares their voice with the voices you've named before, and that comparison works best with voices measured by its own speaker separation, the same way it measures them for its own transcripts. A service's speaker labels can also merge two people into one speaker, or split one person into two; separating the voices on your Mac avoids that, so the people you know come back with their names more often, and you have less to fix in Identify Speakers.

What happens, step by step:

1. As soon as the service has the audio, your Mac starts separating the voices, using its own copy of the recording, so it's usually finished before the service is.
2. When the service's transcript arrives, each line goes to the speaker your Mac heard most during that line.
3. Voice matching names the people it recognizes. A name the service guessed (from someone saying "Thanks, Dana", say) is kept as a suggestion for any speaker no voice matched.

Nothing is sent anywhere for this step. If it fails, or finds no voices, the service's own speakers are kept. A meeting where you've already confirmed someone is never re-separated. It applies to new transcripts; meetings already in your library keep their speakers. One limit: a service writes in lines, not single words, so a line where two people talk over each other goes to whoever spoke most of it.

You can switch it off in Settings → Speakers → Customization… (Separate speakers on this Mac for Plaud, Pocket and Fireflies transcripts). Then the service's own speakers are used and your Mac does no work for these meetings, but voices are recognized less reliably.

When a meeting is transcribed on your Mac as well: only if the service couldn't make a transcript at all (it was unavailable, refused the file, or found no speech, as Plaud does for recordings without talking). Then MacParsley transcribes it on your Mac instead, with a message saying why, so you're never left without a transcript.

How sure is it?

Names are shown with a color:

You can change those thresholds, and how eagerly similar voices are split apart, in Settings → Speakers → Customization….

Naming people: Identify Speakers

Open Identify Speakers… (a button on the Audio, Transcript and Summary tabs; it reads Speakers… once everyone is named).

1. At the top it shows the meeting's title, date and length, so you know which meeting you're in.
2. Choose a speaker from the menu at the top. Their lines are listed; click one to hear it.
3. Type the person's name. Known people appear as you type; choose one, or choose to create a new person. If a suggestion is yellow, click Confirm.
4. If some lines belong to someone else, tick them and move them to another speaker.
5. Press Done to apply (or Cancel to discard). Nothing changes for good until Done.

Confirming a name teaches MacParsley that voice, so the next meeting names that person by itself.

After an import of many recordings, the Unidentified Speakers smart folder lists the meetings that still need you. See Importing.

Managing people

Settings → Speakers lists everyone MacParsley knows, with a search box and how many voice prints each has ("2 of 12 voice prints"). Edit a person to rename them, link them to a contact in your Contacts, or delete them (their voice prints are deleted and their meetings go back to showing Speaker 1, Speaker 2…).

Settings → General → Voice Prints → Keep per speaker (4 to 128, default 12) sets how many prints are kept per person. It's a rolling set: once a person has that many, a clearer print replaces their weakest.

Two different people with the same name are kept apart if you choose Create a Separate Speaker; otherwise merge them.

Fixing misheard words

Speech recognition gets names and jargon wrong. To fix a word:

1. Select it in the transcript (or in a summary) and right-click → Correct Transcription….
2. Type what it should say. Two switches control how far it reaches:
   • Also fix this in future transcripts, summaries and titles (on by default) remembers the fix.
   • Only where the capital letters match makes matching case-sensitive. Otherwise capital letters, spacing, plurals and "'s" endings don't matter: Chord Killers → Cordkillers also fixes chord killer, Chordkillers and Chordkiller's.
3. Correct changes every place in this meeting: the transcript, the summary, the title and the mind map. If the old wording also appears in other meetings, MacParsley lists them with check boxes (red for what changes, green for the new wording), with Select All, so you can fix those too in one step.

The remembered fixes are listed under Settings → Summaries → Corrected Spellings, where you can add, change and remove them. Think before adding a short word: a fix like DIM → DIMM would also change the ordinary word "dim" wherever it's transcribed.

Corrections only change text MacParsley already has. They don't change the audio.

← The Recording tabSummaries and mind maps →