How to Record Audio for Better Transcripts

A clean transcript starts with a clean recording. Put the microphone close to each speaker, record in a quiet room, have people speak one at a time, and do a short test before you start. Then upload the audio track, set the language, and tick Label different speakers if more than one person talks.

Transcription tools turn speech into text. They work best when the speech is the loudest, clearest thing in the file. Every other sound, from an air conditioner to a second conversation, is something the model has to guess around. Most of the mistakes people blame on transcription software are decided before the file is ever uploaded, at the moment of recording. The good news is that the fixes are cheap. Almost all of them are about where you put the microphone and how people speak, not about buying equipment.

This guide applies to every kind of recording people bring to Kai The Scribe: meetings, interviews, podcasts, lectures, voice memos and video narration.

1. Get the microphone close

Distance is the biggest factor you control. The further a microphone is from a mouth, the more it records the room instead of the voice: echo off walls, the hum of electronics, chairs moving. A phone or laptop microphone across a table is the most common reason a transcript comes back patchy.

• Use an external or clip-on microphone when you can. Built-in microphones on phones, laptops and cameras are designed to pick up everything around them, including the device itself.
• Keep it close but not touching. A hand-width or two from the mouth is a good start. Too close and breaths and hard consonants pop; too far and the room takes over. Try a few distances in your test.
• One microphone per person is ideal for interviews and podcasts. If you only have one, place it between the speakers, nearer the quieter voice.

2. Choose a quiet room

Background noise competes with speech. Music, traffic, a busy café and the hum of a fan all make it harder to pick out words.

• Turn off fans, air conditioning and notifications for the length of the recording.
• Pick a room with soft furnishings, such as curtains, carpet and sofas. Bare walls and hard floors bounce sound around and add echo.
• Close windows and doors, and put a note on the door if people might walk in.
• On calls, ask everyone to mute when they are not speaking and to wear headphones to stop echo.

3. One voice at a time

Crosstalk, where two people talk at once, is the hardest thing for any transcription tool to separate. It also confuses speaker labelling, because the words from two voices land in the same moment.

• Let each speaker finish before the next one starts. A short pause between turns helps.
• Ask people to speak at a steady pace and to face the microphone.
• Say names, product terms and acronyms clearly, and spell unusual ones out loud at the start. You will need the spellings when you check the transcript.

4. Always record a test first

Record ten seconds of every voice, then play it back on headphones. Listen for a muted or wrong microphone, a hum, one voice much quieter than the others, and clipping, the harsh crackle when a level is set too high. A test takes less than a minute and catches the problems that would otherwise cost you the whole session. Do one each time you use new equipment or a new room.

5. Save and upload the right file

Kai The Scribe accepts MP3, WAV, M4A, MP4, MOV and WEBM files. A few choices make the upload smoother:

• Upload audio, not video, for long recordings. The audio track of a meeting or lecture is a fraction of the size of the video. Uploads are capped at 100 MB. Zoom, for example, saves an audio_only.m4a next to every local recording; the Zoom guide shows where to find it.
• Use a direct link for big files. If a file is too large to upload, put it somewhere public and paste a direct link to the file under Media URL.
• Do not over-process. Heavy noise reduction or compression can make speech sound smeared. Light cleanup is fine; the best fix is still a better recording.

6. Set the options to match the recording

• Language. Auto Detect works for most files. When you know the language, pick it from the menu, which lists English, Spanish, French, German, Italian, Portuguese, Dutch, Polish, Russian, Japanese, Korean and Chinese.
• Label different speakers. Tick it for any recording with more than one voice. The transcript then starts each turn with Speaker 1:, Speaker 2: and so on. Leave it off for a solo recording.

After transcription: check what matters

Even a clean recording produces a draft, not a final record. Names, numbers and technical terms are where any transcript is most likely to slip. Search for them, fix the spelling once with find and replace, and play back any line you plan to quote or publish. For step-by-step guides by recording type, see transcribing an interview, transcribing a lecture, and podcast transcription.

Frequently asked questions

Transcribe a recording