Give the transcription clean audio
Accuracy starts at the microphone. Speech recognition copes well with accents and fast talkers, much less with two people speaking at once, echoing rooms and music under the voice. If you have a version of the episode without the music bed, transcribe that one.
Upload the best file you have. A WAV or a high-bitrate MP3 gives slightly better results than a file that has been compressed several times. Video works too: drop the MP4 and the audio track is extracted automatically.
Multitrack recordings, with each guest on their own track, are ideal for editing. For the transcript, use the final mix: it is what listeners hear, and the timestamps will match the published episode.
Upload a file or import from your feed
For a single episode, drop the file. Recordings up to 2 hours are accepted; split anything longer, such as a full-day event, into parts.
For a back catalog, paste your podcast’s RSS feed address. Episodes appear with their titles and dates, and you tick the ones to transcribe. Only the audio files listed in the feed are fetched; nothing in your hosting account is touched.
Each minute of audio uses one credit. A 45-minute episode uses 45 credits, so the free plan suits short shows and tests, while the 1,000 monthly credits of Pro cover a weekly one-hour show several times over.
Set the language and the speakers
Choose the spoken language rather than relying on detection when you know it. Detection listens to the opening minute, and an English intro on a Spanish show can mislead it. More than 40 languages are supported.
Set the number of speakers if you know it. Telling voices apart is more reliable with the right count, especially when two voices sound alike, such as two hosts of the same age and accent.
Add the names and terms to spell correctly: guests, brands, places, jargon. Each one is then written the way you typed it instead of being guessed from its sound, so a guest’s surname comes out right every time instead of in five different spellings.
Review the transcript efficiently
Even an accurate transcript has a few errors per page, and they cluster around proper names, numbers, crosstalk and words said while laughing. Review with the audio: click any word to play from that point.
Rename each speaker once, so that “Speaker 1” becomes “Maya”, and the name changes everywhere. Use search and replace for a word misheard the same way throughout.
Choose a style. Clean verbatim removes “um”, false starts and repetitions, which most readers prefer; full verbatim keeps them, which matters for research interviews or legal records. Clean verbatim is the default.
Export and publish
TXT is plain text for your website or a newsletter. DOCX keeps speaker names and timestamps in a document you can edit and share. For video platforms and players, export SRT or VTT subtitle files; the subtitles guide explains the difference.
Publish the transcript on the episode page, not only as a download. Search engines can read the text, listeners who are deaf or hard of hearing can follow the conversation, and newcomers can skim before deciding to press play.
Keep the transcript at hand: it is the raw material for show notes, chapters, quotes for social posts and the episode description.
Frequently asked questions
How accurate is automatic transcription?
On clear speech with one person talking at a time, expect a few corrections per page, mostly names and jargon. Crosstalk, loud music and poor microphones lower accuracy; the list of names and terms helps most.
How long does it take?
Usually a few minutes for a one-hour episode. You can close the tab: you get an email when the transcript is ready.
Can I transcribe a video?
Yes. Upload an MP4 and the audio track is extracted automatically. The same 2-hour limit applies.
Which languages are supported?
More than 40, including English, French, German, Spanish and Arabic. Choose the language before you start for the best result.
