Whisper
Upload a recording and get the full transcript back.
Audio Transcription: Turn Speech into Text with Word-Level Timestamps
Upload a recording and get the full transcript back, plus the exact start and end time of every single word. Interviews, podcasts, voice memos, lectures: if someone’s talking in it, this turns it into text you can search, quote, subtitle, or feed into your next step.
It runs on OpenAI’s Whisper, trained on 680,000 hours of speech across roughly 100 languages. You don’t need to tell it what language the audio is in; it figures that out on its own.
What can you do with it?
- Make subtitles and captions with frame-accurate timing, since every word carries its own timestamps
- Transcribe interviews and meetings instead of scrubbing through the recording for that one quote
- Turn podcasts into text for show notes, blog posts, or a searchable archive
- Caption your videos by chaining it with a subtitle step in a workflow
- Summarize long recordings by piping the transcript into an LLM step
- Transcribe in ~100 languages, from English and Spanish to Turkish and Japanese
How to use it
- Upload your recording. Any common audio format works (mp3, wav, m4a, flac), up to 25MB.
- Optionally set the hints. Pick the spoken language if you know it, and add names or jargon in the hint field so they come out spelled right (“Mitte”, “Kubernetes”, “Dr. Okonkwo”).
- Run it. A typical recording transcribes in well under a minute.
Inside a workflow you can skip the upload entirely: the transcription step picks up the audio produced by the step before it, like a voice generator or a stem splitter’s vocal track.
What you get
Structured JSON, ready for downstream steps:
- Text: the full transcript in one string
- Words: every word with its start and end time, in seconds
- Segments: sentence-level chunks with their own timestamps, handy for subtitle lines
- Language: what Whisper detected the audio to be
- Duration: the length of the recording
Tips
- Feed it clean audio. Background music and crosstalk hurt accuracy; if your source is a busy mix, split the vocals out first with the Stem Splitter.
- Use the hint field for unusual words. Product names, acronyms, and people’s names transcribe far better when Whisper has seen the correct spelling up front.
- Set the language for short clips. Auto-detect is reliable on longer audio, but a 5-second clip gives it little to work with.
- Word timestamps beat segment timestamps for karaoke-style captions. Use segments for classic subtitle lines, words when each one should light up as it’s spoken.