Yes, you can transcribe a video to text for free, but the free tier is small: one run a day of up to two minutes for a visitor, and five credits a day with an account. The recogniser handles English and simplified Chinese, and the file is uploaded for processing rather than transcribed on your device.
A run costs five credits per minute of media, rounded up, so an account covers about one minute of speech a day and a two minute clip spends ten credits. Paid plans raise both the daily credits and the length ceiling, and none of them change the per minute price. The two limits that surprise people are language and upload size. There is no French, Spanish, German or Japanese option, and every plan is capped at a 30 MB upload, which a long 1080p video will hit before it hits the duration limit.
Search for a free transcription tool and the results all promise the same thing: upload a video, get the words back. The differences are in what the words come back attached to, how much of them you get before a paywall, and what happens to the file while it is being worked on. This page covers all three, including the cases where this tool is the wrong choice.
Video to Subtitles
Speech recognition writes the text with timecodes. You fix it in the preview, then download SRT, VTT or TXT. The file is uploaded for this one.
What "transcribe to text" actually produces
A transcript is not one thing. The same recognition pass can be packaged three ways, and picking the wrong one is the most common reason people redo a job.
| Format | What is inside | Where it goes |
|---|---|---|
| SRT | Numbered lines with start and end timecodes | Uploaded as captions, or burned into the picture |
| VTT | The same cues in WebVTT syntax | An HTML video element or a web player |
| TXT | Spoken words in reading order, no timecodes | Notes, a document, a blog draft, a search index |
SRT is the default because almost every player and platform accepts it. VTT is what a web player expects, and converting between the two is a text operation rather than a re-run. TXT is the right answer when the goal was never captions: for a webinar summary, the timecodes are noise.
When speaker labelling is switched on, each line is tagged with the person who said it, numbered from one upward. It is worth having for an interview and wasted on a solo recording.
How to transcribe a video in the browser
- Add the file. Drop a video or audio file up to 30 MB. Accepted audio formats are MP3, WAV, M4A, AAC, FLAC and OGG, and accepted video formats are MP4, MOV, MKV, WebM, M4V and MPEG. If the clip is long, extract the audio first, because a small audio file is far more likely to fit under the upload cap than the video it came from.
- Tell it the language, then start. Pick English or Chinese if you know which one it is, or leave it on automatic. Turn on speaker labels only if more than one person talks. The page shows elapsed time while the run is queued and processing, so you can leave the tab alone instead of watching a spinner.
- Fix the text in the preview, then download. The preview shows the whole transcript with timecodes and it is editable. Names, product names and numbers are the usual errors, so read once with the audio playing. The download buttons export whatever is currently in the box, which means your corrections are what lands in the file.
What decides whether the transcript is usable
Automatic speech recognition is not uniformly good or uniformly bad. It is predictable, and the prediction depends on the recording rather than on the brand on the website.
| Recording | What to expect |
|---|---|
| One speaker, close microphone, quiet room | Near publishable after a read through |
| Two speakers taking turns, decent audio | Good, with the odd missed word at the handover |
| Several people talking over each other | Lines merge and attribution becomes guesswork |
| Music or traffic louder than the voice | Gaps and invented words where the voice is masked |
| Laptop microphone across a meeting room | Usable outline, heavy editing needed |
| Strong regional accent or technical jargon | Terms and names wrong, ordinary words fine |
The pattern behind that table is signal to noise. Recognition does not need loud audio, it needs clear audio, and a quiet phone recording held near the mouth beats a loud room recording every time. If you control the recording, the cheapest fix is a microphone within twenty centimetres of the speaker and no music underneath.
If you do not control the recording, cut a noisy intro and level the audio before transcribing. The audio trimmer does both locally.
The free allowance in plain numbers
Free transcription offers usually hide the limit until after the upload. Here it is before you start.
| Who | Daily allowance | Longest single run |
|---|---|---|
| Visitor, no account | One run per day | 2 minutes |
| Free account | 5 credits per day | 10 minutes |
| Paid plan | 1500 credits per month | 30 minutes |
Credits are spent at five per minute of media, rounded up, so a 61 second clip costs ten. Five credits a day therefore buys about one minute of speech, which is enough to test the tool on a real recording and not much more.
Two rules about when credits are taken matter more than the size of the allowance. Credits are spent once the file has been processed, so a clip that turns out to contain only music, or only silence, still costs its length. A run that fails before processing, or that is rejected for being too long or in the wrong format, costs nothing. Repeating a poll on a finished job never charges twice.
The 30 MB ceiling, and why it arrives first
Every plan shares the same upload limit of 30 MB, and video is a heavy format. A minute of 1080p footage is often 50 to 150 MB depending on the bitrate, so the size limit can stop a clip that is well inside the ten minute duration limit.
The fix is to stop sending video at all. Speech recognition only needs the audio track, and the audio track of that same clip is a fraction of the size. Extract it, then upload the audio file. Our video to MP3 tool does the extraction in the browser with no upload, and a mono MP3 of a ten minute talk is usually a few megabytes, which leaves the duration limit as the only thing to think about.
If a recording is longer than one run allows, split it into parts. Each part starts its timecodes at zero, so you add the offset yourself when you stitch them, and there is no batch mode.
This one uploads the file
The rest of this site is built so that files never leave the browser. This tool is the exception, and it is worth being explicit about why.
| Video to Subtitles | The other 45 tools | |
|---|---|---|
| Where the file goes | Uploaded, then forwarded to a speech service | Stays in the browser tab |
| Works offline | No | Yes, after the first load |
| Cost | Credits per minute | Free and unlimited |
| Model size | Large server model | Small in browser engines |
A server side recogniser is more accurate than anything that fits in a browser tab, which is why this job is not done locally. The price is that the recording leaves your device.
Recorded a client call under an NDA, a medical consultation, or a draft of something unreleased? Do not upload it. Use a local program, accept the slower setup, and keep the file on disk. Recorded a public talk, a product demo you are about to publish anyway, or your own voice reading a script? The upload is the better trade.
What this tool cannot do
These are the limits that show up after the first run, so they are worth reading first.
- It does not translate. English speech produces English text. There is no path from an English recording to Chinese subtitles in one step; you get the transcript, then translate it somewhere else.
- It does not cover most languages. English and simplified Chinese only. Other languages complete the run, spend the credits, and return unusable text.
- It does not align to individual words. Timecodes mark cue boundaries, not karaoke style word highlighting. If you need word level timing for a social caption effect, this is not that tool.
- It does not separate speakers automatically into named people. Labels are numbers, so Speaker 1 and Speaker 2, and you rename them in the preview.
- It does not process a folder. One file per run, one run at a time.
- It does not improve a bad recording. Noise, crosstalk and distance are baked into the audio by the time the recogniser sees them.
- It does not keep the media. No video download link is produced; only the text comes back.
After the transcript: two ways to finish the job
Once the text exists, the rest of the work can be done without uploading anything, which is the useful part of this workflow.
If the goal is captions on the video
Download the SRT, then open the add subtitles tool. It reads the file, lets you set font, size, colour and position, and burns the text into a new video in the browser. No watermark, no upload, and the styling is yours rather than a fixed preset.
If the goal is text you can read or search
Download the TXT and stop there, or download the SRT and convert it. The subtitle converter turns SRT into VTT or ASS and can shift the whole timeline by a fixed offset, which is the fastest fix when every cue sits two seconds early.
That three step chain, recognition then conversion then burning, is why this tool exists as a separate page rather than as a checkbox inside the subtitle editor.