Before you record
A clean recording saves more time than any editing trick. Ask for permission to record, and follow the consent rules that apply where you and the other person are. Then set up for two clear voices:
- Choose a quiet room. Turn off fans and music, and keep away from traffic and other conversations.
- Keep the microphone close to both people. A phone on the table between you works; one person's laptop microphone across the room does not.
- Record a short test and listen back before the real conversation.
- Write down names and terms as the guest says them: people, companies, places and jargon. You will need the correct spellings later.
- For a remote interview, record the call, or ask each person to record their own side as well.
Transcribe the recording
Open the transcription studio, choose the recording and pick its spoken language, or let StemVocal detect it. Audio files such as MP3, M4A and WAV work, and so do video files: an MP4 or MOV is transcribed from its sound.
For a one-on-one interview, set speakers to Up to 2 so the transcript separates the interviewer from the guest. Speaker labels use credits. A free transcript has no labels; when it finishes, Add speaker labels takes you back to the file's options with labels chosen and charges nothing until you press Transcribe. Rename the labels to "Interviewer" and the guest's name, and every turn follows.
Free jobs cover recordings up to 30 minutes. A longer interview, up to 3 hours, uses one credit for each started 30 minutes. If you have several interviews, choose them together to make a batch of up to 20 files and download every transcript in one ZIP.
Check every quote
An automatic transcript is a first draft, and a quote has to be exact. Before you use one:
- Click the first word of the quote to hear it in the recording.
- Slow the playback down, from 0.5× upward, for fast or unclear passages.
- Correct the text where it differs. Your edits save as you go.
- Search the transcript for the other places the same topic comes up, so a quote is not taken out of context.
Names and numbers deserve a second listen: digits, dates and amounts are easy to mishear. When people talk over each other, the transcript usually keeps one voice, so interruptions are where words go missing. For a name misheard the same way throughout, Find & replace fixes every instance, and one Undo restores them all.
Verbatim or clean?
Decide how literal the transcript should be before you edit. A verbatim transcript keeps false starts, repeated words and fillers such as "um"; a clean transcript removes them while keeping the meaning, which reads better in an article. Many research projects ask for verbatim text, while articles usually quote cleanly without changing the meaning. An automatic transcript may not follow either style exactly, so read it against the audio and edit to the one you need.
Export for writing and citing
| Next step | Download |
|---|---|
| Write the article or report | Word, with a timestamp and speaker at the start of each turn |
| Paste notes into another app | Plain text, or Copy text |
| Code answers in a spreadsheet | CSV, with start and end seconds and the speaker for each turn |
| Share a fixed copy | Print / PDF, through your browser's print dialog |
Interviews can hold personal details. When a recording is uploaded for transcription, it goes to private storage with access for 24 hours and deletion normally within another 24 hours, and you can delete it sooner from the editor; a free transcription that runs in your browser keeps the recording on your device. Your edited transcript saves in this browser under Saved transcripts; download a copy of anything you need to keep. Recording a podcast episode instead? See podcast transcription.
Good to know
How long can the interview be?
Up to 30 minutes with a free job; you get 2 free StemVocal jobs each UTC day without an account. With credits, files up to 3 hours and 1 GB work, at one credit per started 30 minutes: a 50-minute interview uses two.
Can it tell the interviewer and the guest apart?
Yes, with speaker labels, which use credits. Choose up to 2 speakers for a one-on-one interview, then rename the labels. They are estimates, so check the places where people interrupt each other.
Which format is best for quoting?
Word with timestamps. Each turn starts with its time and speaker, so you can return to the exact moment and confirm a quote before you publish it.
Is this a certified transcript?
No. It is an automatic transcript that you correct yourself, not a certified or human transcription service. Check names, numbers and quotations against the recording.