How to Transcribe Audio to Text for Free (Without Uploading It)

To transcribe audio to text for free, open Transcribe Audio, drop your audio or video file (or record from your microphone), pick a model, and press Transcribe. Lines appear with timestamps as each part is done. Click any line to fix a word, then copy the text or download it as TXT, SRT or VTT.

The speech recognition runs inside your browser using Whisper, an open model released by OpenAI. Your recording is never uploaded. The only thing downloaded is the model itself, the first time you use it.

Why transcribe on your own device?

Most online transcription services upload your file to their servers, run it there, and keep it for some period of time. That's fine for a podcast that's going public anyway, and much less fine for:

• a job interview or a doctor's appointment you recorded,
• a client call covered by an NDA,
• a lecture you're not supposed to share,
• personal voice notes.

When the model runs in the browser, the audio stays on your laptop or phone. There's no account, no "free minutes" counter and no waiting in a queue, because your own device is doing the work. The trade-off is speed: a server with a big graphics card is faster than an old phone. On a recent laptop with WebGPU (Chrome and Edge have it), the difference is small.

Picking a model: fast, better or best

Whisper comes in sizes. Bigger models are more accurate but take longer to download and run.

Model	Download (CPU)	Good for
Fast (tiny)	about 40 MB	Clear speech, quick drafts, slow devices
Better (base)	about 77 MB	Most recordings: the best balance
Best (small)	about 250 MB	Accents, noise, several speakers, other languages

Each model is downloaded once and then kept in your browser's cache, so the second time it starts right away. If your browser has a usable GPU, the tool runs the model there. The GPU files are a bit bigger, but transcription is many times faster.

A simple rule: start with Better. If you see wrong words in places where the speech is clear, switch to Best. If your phone struggles, use Fast.

Getting the best accuracy

Whisper was trained on hundreds of thousands of hours of real-world audio, so it copes well with ordinary recordings. These habits still make a big difference:

1. Record close to the speaker. A phone 30 cm from someone's mouth beats a laptop mic across the room every time.
2. Cut the music. Background music is the single biggest cause of missed words. If you're transcribing a video, a section with music under the voice will be weaker.
3. Pick the language yourself for short clips. Automatic detection listens to the first 30 seconds. For a 5-second clip, or audio that starts with music, choosing the language yourself is more reliable.
4. Use Translate to English only when you want English. Otherwise leave it off, so you get the speaker's own words.
5. Proofread names and numbers. Speech models are good at ordinary words and worst at names, brands and figures. The editable transcript is there for exactly that.

Long recordings

Whisper listens to 30 seconds at a time. For longer files, the tool cuts the audio into parts at the quietest moment near each 30-second mark, so a word is rarely split in half, and then stitches the timestamps back together. You see each part's lines as soon as it's done, plus an estimate of the time left.

The limit is 60 minutes per file, because the whole recording has to sit in your browser's memory while it's processed. For a two-hour meeting, split it first with Trim Video. If you only have a video and want a smaller file, Video to MP3 keeps just the sound.

Turning a transcript into subtitles

Every line in the transcript has a start and end time, which is all a subtitle file needs:

• SRT is the classic format. Each block has a number, a time range like 00:01:02,500 --> 00:01:05,000, and the text. Almost every video editor and player reads it.
• VTT (WebVTT) is the web's format, used by HTML5 video players. It looks almost the same, with dots instead of commas in the times and a WEBVTT header.

Fix any wrong words before downloading, because your edits go straight into the file. Then upload the SRT to YouTube under Subtitles, or import it into your editor. If you're writing a caption for the post too, the caption and hashtag generator can take the transcript as its starting point.

Recording straight from the microphone

You don't need a file. Press Record from your microphone, talk, press Stop, and the recording goes straight into the transcriber. It's handy for dictating notes, and for checking how your own voice comes out before you record something longer. If you want to keep the audio file as well, use the Voice Recorder first, then drop the saved file in.

Common problems

• "This file couldn't be read." Your browser can't decode that format, or the file has no sound track. Convert it to MP3 and try again.
• It's very slow. You're probably on the CPU. Pick the Fast model, close other tabs and keep the screen on (phones pause background work).
• Out of memory. Use a smaller model or a shorter file. Phones have much less memory available to a web page than laptops.
• Repeated or invented sentences in silent parts. This is a known Whisper quirk in long silences or music. Silent parts are skipped automatically, but check any line that doesn't match what you hear: click its time to play that moment.