Updated on
Where the captions come from
The words in a clip come from the transcript of your episode, with the time each word is spoken. You make a transcript on the tab Audio of your episode. It takes a few minutes, and transcriptions are part of every plan from Basic. You can also upload your own transcript there.
Without a transcript
You can still make a clip. Pick your moment by time, and we transcribe just that part for the captions. Choosing in the text and suggestions by AI are not available until there is a transcript, because without one nobody knows where a sentence sits in the audio.
An uploaded transcript without timings, such as a plain text file, counts the same as no transcript for clips. Have the transcript made in PodcastFeed, or upload an SRT or VTT file with times.
What AI does, and what it does not
- Transcribing is done by Voxtral, a model by Mistral. Your audio goes there to be turned into text.
- Suggesting moments is done by Mistral as well. Only the text of your episode is sent, never your audio. Suggestions are part of Standard and higher.
Mistral is a European company. For every clip, the audio of just the part you picked goes to Voxtral once more, to time each word to the voice. That happens with or without a transcript. Suggesting moments is the one step you choose to use or not: picking a moment yourself works in every plan.
Fixing a word
Is a name spelled wrong in your captions? Correct it in the transcript on the tab Audio and make the clip again. Your word library applies to new transcripts, so a name you corrected once comes out right the next time.