How to transcribe a recording: Six steps and the transcription rules

Transcribing a recording in six steps

1. Give it the file. On New transcript, choose File and drop the recording. MP3, M4A, WAV and common phone formats open directly; MP3 to text has the full list.
2. Check the language. With the wrong one, the text turns to nonsense.
3. Get the draft. Press Prepare the model, then Start transcription when it is ready. By default the sound is processed on your device.
4. Correct as it comes. When you see a wrong name, click and type. Editing while it runs does not disturb the new text.
5. Compare with the audio. Clicking a paragraph’s time takes the recording there. Always listen to numbers, names and quotes.
6. Format and download. Apply the rules below and use Export for a DOCX or TXT.

By hand or automatic? Counting the time

To transcribe audio by hand takes several times the recording’s length, even for someone practised. With a tool the typing goes away and listening and correcting remain.

A mix is often best: The machine writes the draft and you finish the hard parts by hand.

Transcription conventions: Punctuation and style

There is no single official rule. What matters is keeping one rule through a whole project. This set works for most jobs:

• Punctuation: Speech does not come in sentences; you divide it. Place commas for sense rather than breath, and put direct speech in quotation marks.
• Cut-off sentences: Leave them with an ellipsis: We were then...
• Repetition: Removed in news and reports, kept in a verbatim transcript.
• Numbers: Choose one rule and keep it. Follow your style guide.
• Names: Confirm the spelling at the source. If unsure, mark it and ask the person.
• Foreign words: Use the correct spelling, not how it sounded.

“Um”, “you know”: When to keep filler words

It depends on the purpose. News and reports drop them because they add nothing for the reader. Language, discourse and some qualitative studies keep them because they carry meaning.

LiveType does not write hesitation sounds like “um” and “hmm” in English and Turkish transcripts. It does not remove words like “you know”, though the model sometimes skips them on its own. If their number matters, do not trust the automatic text: Check by ear.

Transcript marks: Laughter, pauses, inaudible parts

Writing the time beside an inaudible part makes it easy to listen again later.

Transcript example: Before and after

The same speech three ways. The automatic draft:

So we you know we said in that meeting the budget comes in March but it didn’t. Mr Brown said the same that it would land on the fifteenth.

Verbatim, corrected by ear:

So we… you know, we said in that meeting the budget comes in March, but it didn’t. [pause] Mr Braun said the same, “It’ll land on the fifteenth.”

Cleaned for news or a report:

We said the budget would come in March, but it did not. Mr Braun also said the money would arrive on the 15th.

What listening caught is “Brown” against “Braun”. You cannot tell that from the text, so always listen to names.

iPhone and Android recordings

iPhone: The Voice Memos app has its own transcription in a list of languages that includes English. For a language outside that list, or to keep working on the text, save the recording to Files from the share menu and choose the M4A file.

Android: Take the file from the recorder’s share menu or copy it to a computer. Old AMR and 3GP formats may not open in a browser; convert them to M4A or MP3 first.

The same page works on a phone, but only small models run there and it is slower. For long recordings plug the device in and keep the screen on, or use a computer.

WhatsApp voice messages

WhatsApp’s own feature makes voice message transcripts on the device. Turn it on under Settings → Chats → Voice message transcripts, choose a language, then press and hold a message to transcribe it (WhatsApp Help). The languages differ by device. If yours is missing, export the message as a file:

1. Press and hold the message. On Android use Share in the three-dot menu; on iPhone use Forward, then the share icon. Menu names may differ by version.
2. Save the file to your device. It will be an OPUS file, which LiveType opens directly.
3. Choose the file on New transcript.

Use messages from others only for your own purposes, and check the rules before you share the text.

Transcription checklist

Before you hand over or use the text:

• The type of transcript was set up front and one rule runs throughout.
• All names of people, organizations and places were confirmed at the source.
• Numbers, dates and amounts were checked by ear.
• Every quoted sentence was compared with the audio.
• Sentences that look unsaid or off topic were checked.
• Speaker labels were replaced with real names.
• [inaudible] marks were listened to again.
• If it will be shared, permission was given or names were removed.

When automatic text is not enough

With very noisy audio, heavy accents and overlapping speech, automatic text is only a draft. The model can mishear a name, miss a “not”, or even write a sentence nobody said: AI transcription accuracy explains why. Do not publish words against a person’s name without checking them by ear. Research needs stricter rules: See qualitative interview transcription.

The free plan has 30 minutes in total. When the text is done, speakers and export covers what comes next.