feemag/ tools
Browse tools

Audio tool · runs in your browser

Transcribe audio and video to text

Turn speech in an audio or video file into text and subtitles. The AI runs in your browser, so your recording is never uploaded. Free, with no account and no minute limit per day.

Loading tool…

How to transcribe audio or video to text

  1. 1. Choose the recording

    Drop an audio file (MP3, M4A, WAV, OGG, FLAC) or a video (MP4, MOV, WebM, MKV). The tool reads its length and tells you roughly how long it will take. Nothing is uploaded.

  2. 2. Transcribe

    Leave Detect automatically on, or choose the spoken language, then select Transcribe. The first time, about 91 MB of AI model downloads and your browser saves it. The text appears in 5-minute parts while you wait; keep the tab open.

  3. 3. Check and download

    Read through the transcript, especially names and numbers, then download it as text (.txt), subtitles (.srt) or WebVTT (.vtt), or copy it.

Make subtitles for a video

The .srt file has numbered lines with start and end times, at most two lines of 42 characters each and seven seconds on screen, which suits most players. In YouTube Studio, open the video, choose Subtitles, add a language and upload the file with timing. In VLC, choose Subtitle, then Add Subtitle File. Video editors such as Premiere Pro and DaVinci Resolve import SRT onto the timeline, where you can fix words and timing.

WebVTT (.vtt) is the same idea for web video players and the HTML <track> element.

What affects accuracy

The tool uses Whisper base, a small open speech model that fits in a browser. Clear speech close to the microphone works best. In our tests with clear, read English, roughly 1 word in 10 differed from the reference text. Accents, background noise, music, phone-call audio and people talking over each other lower accuracy, and languages other than English usually need more correction.

Speech models can also invent text over long silence or music. The tool removes phrases repeated many times in a row, but read those parts yourself.

Interviews, lectures and meetings

The transcript starts a new paragraph at longer pauses, which roughly follows changes of speaker or topic. It does not label speakers, so add names after you download. For a recording longer than 60 minutes (20 on a phone), split it into parts and transcribe them one after another.

Why the recording is not uploaded

The AI model downloads from Hugging Face and its runtime from this site, and your browser does the work. The audio is decoded, turned into text and saved as files inside this page. No audio, file name or text is sent to Feemag or anyone else. That makes it suitable for recordings you would rather not hand to a transcription service, such as interviews, medical appointments you recorded for yourself, or internal meetings.

If it is slow or stops

Transcribing takes roughly half the length of the recording on a recent laptop and longer on phones. Other heavy tabs slow it down. If the browser runs out of memory, close other tabs or split the file. If your browser cannot read the file, convert it to MP3 or M4A and try again.

Built by Faheem Riaz at Feemag. How file privacy works · Report a problem

Good to know

Questions & answers

Is my recording uploaded?

No. The AI model downloads to your browser once and the audio is turned into text on your device. The recording, its name and the transcript are never sent to Feemag or anyone else.

How accurate is the transcript?

It uses Whisper base, a small open speech model that fits in a browser. In our tests with clear, read English, roughly 1 word in 10 differed from the reference text, counting small differences such as Mr. and mister. Accents, background noise, music, several people speaking at once and languages other than English lower accuracy. Check names, numbers and technical words before you use the text.

Which languages does it understand?

Whisper was trained on about 99 languages, and the menu lists 25 common ones. It works best for English and widely spoken European languages; for many others the text needs more correction. Leave Detect automatically on, or choose the language if the recording starts with music or silence. The transcript is not translated.

How long can the recording be?

Up to 60 minutes per file on a computer and 20 minutes on a phone, and up to 2 GB. For a longer recording, split it into parts and transcribe them one after another.

How long does it take?

On a recent laptop, about half as long as the recording: in our test, 5 minutes of speech took about 2 minutes on an Apple M1. Phones and older computers are slower. The first time, about 91 MB of AI model and runtime downloads; your browser saves it for next time.

What is the difference between TXT, SRT and VTT?

TXT is the plain transcript in paragraphs. SRT and WebVTT (VTT) are subtitle files: short lines of text with start and end times. Most video editors, players and YouTube accept SRT; WebVTT is the format web video players use.

How do I add the subtitles to a video?

In YouTube Studio, open the video, choose Subtitles and upload the .srt file with timing. In VLC, choose Subtitle, then Add Subtitle File. Most video editors, including Premiere Pro, DaVinci Resolve and CapCut on desktop, can import an SRT file onto the timeline.

Can it transcribe a video file?

Yes. It reads the sound from MP4, MOV, WebM and MKV files, as well as MP3, M4A, WAV, OGG and FLAC audio. The video itself is not changed. If your browser cannot read a file, convert it to MP3 or M4A first.

Does it label who is speaking?

No. The transcript is one block of text in paragraphs, without speaker names. Add speaker labels yourself after you download it.

Why does the transcript contain words nobody said?

Speech models can invent text over long silence or music, and sometimes repeat a phrase. The tool removes phrases repeated many times in a row, but read the transcript where the recording is quiet or has music.

How is this different from Otter, Rev or Descript?

Those services upload your recording to their servers and limit, or charge for, the minutes you transcribe. Here the recording stays on your device and there is no account or minute limit. Their larger server models are usually more accurate, especially for noisy audio, and they offer features such as speaker names that this tool does not.

Is it really free?

Yes. Every Feemag tool is free, with no sign-up, no watermarks added by us and no daily limits.

Are my files uploaded anywhere?

No. Your files are processed by your own browser on your device and never uploaded to a server. Processing speed depends on your device and file size.

Does it work on my phone?

Yes: in any modern browser (Chrome, Safari, Edge or Firefox) on iPhone, Android, Mac, Windows and Linux.

More free tools