Transcribe an Interview Free: Record, Convert, Quote Right

Record the interview so transcription is easy

Accuracy is decided before you press record. Whisper handles accents well, but it cannot recover words buried under an espresso machine or two people talking at once.

• In person: a phone voice memo app works if the phone sits on a soft surface about a foot from the speaker, not in the middle of a noisy table. A lapel microphone for each person is better still.
• Remote: record through the call app or a recording service, and ask the guest to use earbuds so your voice does not echo back through their microphone.
• Test first: record ten seconds, play it back, and check that both voices are clear and neither clips.
• Say the basics aloud: the date, the interviewee's name and its spelling, and their consent to be recorded. That gives you a timestamped record and a spelling reference.

Recording consent rules vary. Many US states allow one-party consent, while others such as California require everyone's agreement, and research ethics boards usually require written consent. Ask on the recording every time.

Transcribe an interview privately in your browser

GrabCast AI Voice-to-Text Studio accepts MP3, M4A, WAV, OGG and many video formats. The model downloads once, then is cached, and your audio is transcribed on your own device with no account and no minute limit.

• Model: Base is the default; Small, about 300 MB, is worth the extra download for accented speech or technical vocabulary. Large-v3 Turbo gives the best accuracy on a recent Chrome or Edge with a capable graphics chip.
• Language: choose it explicitly for bilingual interviews so the model does not switch languages mid-answer.
• Progress: long recordings are processed in roughly one-minute pieces with a live transcript and a time-remaining estimate, so you can start reading before it ends.
• Exports: plain text, text with timestamps, SRT and VTT, or a Word document for editors who work in track changes.

If a phone recording will not load, convert it to MP3 or M4A with the Audio Converter first. Voice memo files occasionally use formats a browser cannot decode.

Verbatim, clean or edited: pick a transcript style

Decide what the transcript is for before you edit, because the right amount of cleanup differs sharply.

• True verbatim keeps every um, false start and pause. Qualitative researchers and legal teams often need it, since hesitation can be data.
• Clean verbatim removes fillers and stutters but keeps the speaker's words and grammar. It suits most journalism and oral history.
• Edited transcripts tighten sentences for readability, which is fine for a published Q&A as long as the source approves or the outlet's standards allow it.
• Anonymized versions replace names, employers and places with codes such as P07, a common requirement for university research.

Whisper tends to drop some filler words on its own and adds punctuation, so its draft sits close to clean verbatim. For true verbatim you will need to listen and add hesitations back in.

Label speakers and check every quote

AI Voice-to-Text Studio does not identify speakers, so labeling is a manual step. The timestamped export makes it quick: questions and answers usually alternate, and each line shows where to listen if you are unsure.

• Use short labels such as Q and A, or initials, and apply them with find and replace where the pattern is regular.
• Search the text for every number, name and date, and confirm each against the audio.
• Mark unclear passages as [inaudible 12:41] rather than guessing; a timestamp lets an editor check it.
• Before publishing a direct quote, play that exact passage once more, even when the transcript reads perfectly.

For summaries, the AI Writer or AI Content Pack can pull key points from the cleaned text, but both send it to a cloud AI, so skip that step for confidential or anonymized interviews.

Step-by-step

Common mistakes to avoid

Pro tips

Frequently asked questions