Transcript Maker (transcription + speaker labels, free)
Drop a recording and this page transcribes it on your device, labels who is speaking and builds a draft transcript. Replay each line to check it, then save it as Word (DOCX), PDF, HWPX or TXT. Nothing is uploaded.
Before you start Copyright and recording rules (tap to read)
- Only upload files you have the right to use. Copying or sharing other people's music, videos or recordings without permission can violate copyright law. FreeSoundTools does not store your files on a server.
- Don't upload secret recordings of conversations you were not part of. Recording other people without consent can be illegal.
Loading the tool...
A transcript is a written record of what was said, in order. This page transcribes your recording with the Whisper speech recognition model, uses a speaker segmentation model (pyannote) to label who said each line, and lays it out as [00:01:23] Speaker 1: text. It uses the same models as our meeting notes tool, but keeps every word instead of summarising.
When to use it
- Interviews, counselling sessions or lectures you want on paper, word for word
- Sharing a meeting as a document, line by line and by speaker
- Making a first draft before a professional transcriber checks it
How to use
- Choose the spoken language and the number of speakers (auto if you are not sure), then drop a recording or record with your microphone.
- When the transcript appears, fill in the date, place and participants on the cover. The file name and length are added for you.
- Rename Speaker 1, 2, 3 to real names, merge speakers that are really one person, or change the speaker of a single line.
- Press ▶ on a line to replay that part and correct the text.
- Save as Word (DOCX), PDF, HWPX or TXT.
Transcript layout
- Cover: the title 'Transcript' and a table with date & time, place, participants, audio file and length
- Body: [hh:mm:ss] Speaker name: what was said
- End: a 'Reviewed by a person □ Reviewer ___ Date ___' box
- The PDF embeds the NanumGothic font, so it opens the same on any device.
Supported formats and limits
- Input: mp3 · m4a · wav · ogg · webm · mp4 · mov and other files your browser can play
- Languages: English, Korean or auto-detect
- Speakers: auto or 1 to 5. Overlapping speech may be given to one speaker.
- The first run downloads the transcription model (large model about 250 MB) and the speaker model (about 2 MB).
FAQ
Is my recording uploaded?
No. Transcription and speaker labelling both run in this browser. Only the model files are downloaded the first time; your recording never leaves the device.
How accurate is it?
It can be wrong. Overlapping speech, background noise, accents and quiet voices make mistakes more likely. Use the ▶ button on each line to replay that part and fix the text or the speaker.
Can I submit it to a court or the police?
What you get is an automatic draft. Depending on where you submit it, it may need to be checked by a professional transcriber, so check their requirements first. The transcript ends with a box for the reviewer and date.
Which formats can I save?
Word (DOCX), PDF, HWPX (the Korean Hangul word processor format) and plain text (TXT). All four include the cover table, the timed lines and the review box.
How do I rename speakers?
Speakers start as Speaker 1, 2, 3 in the order they first talk. Type a name in the speaker name box to change it everywhere, merge two speakers that are really one person, or change the speaker of a single line.
Does it work for long recordings?
Yes, but it runs on your device, so it takes time. In our test a 60-minute recording took about 40 minutes on a PC with the medium model and speaker labels. Split recordings over 2 hours if you can.
Open source used
Whisper (MIT), pyannote segmentation model (MIT), transformers.js (Apache-2.0), fflate (MIT), pdf-lib (MIT), fontkit (MIT), NanumGothic (SIL OFL 1.1). The HWPX file is written by our own code. Details are on the open source and model notices page.
Learn more
For summarised meeting notes use AI Meeting Notes, for subtitle files use Subtitle Generator, and for plain text without speakers use Audio to Text Converter.