An interview is only as useful as your ability to find what was said. A good transcript lets you search for a phrase, pull an exact quote and check a fact in seconds, instead of scrubbing through an hour of audio. But a transcript is only as good as the recording behind it, and that is decided long before you press any button.
This guide covers the whole path: how to record so that the text comes out cleanly, how to transcribe the file, how to label speakers and check quotes, and what to do with long or multilingual interviews. It is written for journalists, researchers, students and recruiters, and everything here works without uploading the recording.
Before you record
Spend two minutes on setup and you will save an hour of fixing later. Choose a quiet room with soft surfaces. Carpet, curtains and furniture soak up echo, while an empty meeting room with glass walls makes every word bounce. Turn off fans, notifications and anything that hums.
Agree on the basics with your guest. Tell them it is being recorded and why, and get their consent in a way you can later show: ask them to say it on the recording, or put it in writing. Laws on recording conversations differ across the US, UK and Canada, and between states and provinces, so if you are unsure, ask for consent every time.
Microphone placement
Distance matters more than equipment. A phone placed close to the speaker usually beats an expensive microphone across a table. Aim to have the microphone about a hand’s width to a forearm’s length from each mouth, and avoid putting it where papers, cups or fingers will rustle against it.
- In person with two people, put the recorder between you on a soft surface, nearer your guest than you. Your own voice is easier to hear anyway.
- On a video call, record each person’s audio separately if your software allows it. One clear track per person beats a mixed room recording.
- On the phone, use speakerphone only in a quiet room, or a headset with a built-in microphone. Hold the phone steady.
- Outdoors, shield the microphone from wind and expect extra errors. Moving indoors for the key part of the interview is often worth it.
Keep the recording going through the whole conversation. Pausing and restarting leaves seams that confuse timing and notes, and most of the best remarks come in the minutes after you think you are finished.
Try it on your own file
Transcribing the recording
Open the interview transcription tool and pick your recording. Audio from a recorder (often WAV, which the WAV to text page covers), a phone or a video call export all work, and video files are fine because only the sound is used. Set the language that was spoken instead of relying on detection when you know it, particularly if the recording opens with small talk in another language. Everything happens in your browser, so a confidential recording stays with you.
Choose a model to suit the job. For a sensitive interview with a lot of names and technical terms, a larger model catches more of them, at the cost of time. Our guide to Fast, Standard and High accuracy explains the trade-off. On a computer, Fast handles up to 3 h, Standard handles up to 2 h, High accuracy handles up to 1 h. On phones it's half of that.
When the transcript opens, OmniaScribe shows timed lines next to a player. That timing is the feature that matters most for interviews: every line is tied to the moment it was said.
Marking who said what
The engine produces a single stream of timed lines. It does not tell voices apart, so it will not label the interviewer and the guest for you. For a two-person interview this is rarely a problem, because questions and answers alternate and the pattern is obvious as you read.
Open Edit and add a short label at the start of each answer, such as “Q:” and “A:” or the person’s initials. Do it as you listen back with the player, and fix names and jargon in the same pass. In a panel or a group interview with several voices, label lines while you listen, because the text alone will not tell you who spoke.
Checking quotes before you use them
Automatic transcripts are excellent for finding the part you need, and risky if you quote them unchecked. Speech models smooth over mumbled words, and a single dropped word, such as “didn’t” heard as “did”, can reverse a meaning. Numbers, dates and proper names are the usual trouble spots.
- Search the transcript for the phrase you want, then click its line to hear the moment in the player.
- Listen to the whole sentence, not just the words you plan to quote, so the quote keeps its context.
- Compare names and numbers with a second source, such as a written document or a follow-up message.
- Write down the timestamp next to every quote. It helps an editor, an ethics board or a colleague find the original later.
Remember that a transcript captures words, not tone. A sarcastic aside or a long pause can be missing from the text, so if the meaning depends on how something was said, listen to it again.
Long and multilingual interviews
A long oral-history interview can run past what a model handles on your device. The workspace tells you before it starts. Split the file at a natural break, such as a pause between topics, and transcribe the parts one after another. Joining the text afterwards is a copy and paste.
For an interview in another language, choose that language, or tick Translate to English to get an English working transcript directly from the speech. Treat a translated quote with extra care. Keep the original-language transcript too, and have a fluent speaker check the wording before you publish a translated quote.
Storing and sharing the transcript
A saved transcript is stored in your browser on that device. For something you must keep, export it. DOCX suits a write-up in Word or Google Docs, TXT suits analysis software, and PDF is a read-only copy to send around. Keep the original recording as well, because the audio is your evidence.
If the recording is sensitive, remember that the Settings screen can delete every saved transcript from the browser when you are done. Models stay installed. If you work on a shared computer, use that option, and export what you need first.