What do speaker labels do?
Pepys Transcribe can detect the different voices in a recording and label each turn as Speaker 1, Speaker 2 and so on, with timestamps. Detection is also called diarization. A recording with fewer than two distinct voices comes back without labels.
Labels are generic. Pepys does not work out who a speaker is. You can rename Speaker 1 to a real name in the transcript, and the rename is applied to the transcript and to exports without changing the original labels.
How much do speaker labels cost?
Speaker labels cost 1.5 credits per minute in total. That is the standard 1 credit per minute for transcription plus 0.5 more credits per minute for speaker detection. They are not free: every minute of a labelled recording is billed at 1.5 credits, and a file's total rounds up to a whole number of credits.
If you turn speaker labels on when you start a transcription, the job is billed at 1.5 credits per minute. If you add them to a finished transcript, Pepys charges the difference, 0.5 credit per minute of the recording, so the total is the same 1.5 credits per minute. You need enough credits for the extra charge before it starts. Length is billed in whole minutes, rounded up.
60 minutes, no speaker labels
- Credits
- 60
60 minutes, speaker labels from the start
- Credits
- 90
60 minutes, labels added to the finished transcript
- Credits
- 60 + 30 = 90
Who can use speaker labels?
Any signed-in account can use speaker labels, including an account that only has its free minutes. No purchase, upgrade or subscription is needed; labelled minutes draw on the account's credits, so the free minutes go further without labels. Signed-out visitors need to sign in or create a free account first. On the API, diarize: true works for any account with enough credits. See errors and limits.
See credits and free minutes for how credits work and pricing for the current packs.
What does a recording need for speaker labels?
Speaker detection works from the audio, so the audio has to be available. That has four consequences.
An uploaded recording is deleted about 30 days after upload. After that, speakers can no longer be detected for it; turn labels on early if you need them. See privacy and data handling.
A transcript made from captions, such as a YouTube transcript, never had audio fetched, so it cannot be labelled. TikTok, Instagram, Facebook, Twitch and Kick links also come back without speaker labels. Upload the file instead if you need them.
A file from a connected account such as Google Drive, OneDrive, Zoom or Telegram needs the connection to be active, so reconnect it if Pepys asks.
What happens to my edits if I add speaker labels later?
Adding labels to a finished transcript re-transcribes the recording from the audio, and the new, labelled transcript replaces the old one. The original stays readable while detection runs, which can take longer than the first transcription.
If you have edited the transcript, those edits are lost when the labelled version replaces it. Pepys asks you to confirm before it goes ahead.
Where else can I turn speaker labels on?
Speaker labels can be requested on the API with diarize: true (create a transcription) and in an automatic Google Drive folder import, where each folder has its own switch (Google Drive). They are not an option on the form for transcribing many recordings at once; add them to each transcript afterwards.
Recordings with several voices are the case they are for, such as interviews, meetings and panels. The speaker diarization tool shows the output on a sample.
What speaker labels do not do
Speaker labels are not an accuracy guarantee. Overlapping speech, similar voices and poor audio can produce wrong attributions, so check the labels against the audio before relying on them.
If the recording includes other people, you are responsible for any notice or consent they are owed before you turn labels on.