How to use Audio & Video to Text
- 1Choose a recording of up to 60 seconds and 40 MB. WAV, MP3, M4A, AAC, OGG, FLAC, MP4, MOV, MKV and WebM are accepted.
- 2Select the spoken language, English or Korean. The menu starts on the language of the page you are reading.
- 3Click Transcribe recording. The audio track is prepared first; on the first run the speech recognition files are downloaded, with progress shown in MB.
- 4Review the transcript. Press a timestamp to hear that segment and edit the text box next to it.
- 5Click Download TXT for plain text or Download SRT for timed subtitles. Your edits are included in both.
Review the text against the recording
Listen for names, dates, numbers and any words missed by the model. Short recordings with one clear speaker work best. Use Audio Cutter to trim a longer audio file, or Subtitle Converter to adjust an exported SRT.
Good uses for a 60-second transcriber
A minute is shorter than a meeting but long enough for a lot of everyday jobs: a voice memo you want as text, a short interview answer you plan to quote, a product clip that needs captions before you post it, or a lecture excerpt you want to search. It is also handy for language learners who want to check what a Korean or English speaker said in a short clip, and for making a first draft of subtitles that you then polish and burn into the video with Add Subtitles to Video.
What we saw when we tested it
We tested with synthesized English speech, because we could generate it reliably on a test machine. The first file was a 17-second MP3 (137 KB) announcing a community garden meeting. From pressing Transcribe to the finished transcript took 13.3 seconds in headless Chromium, including about 80 MB of speech recognition files downloaded on that first run. The result came back as three timestamped segments and was almost word-perfect: Saturday, March fourteenth became Saturday, March 14, and the only miss was a dropped a in a reusable water bottle.
The second file was a 35-second, 1280 by 720 MP4 video of a photography lesson (13.6 MB). It took 22.4 seconds in total and returned five segments. Numbers were written as digits (one 500th of a second, F16), and one word was misheard: tripod came out as dry pod. The first segment ran for 21 seconds, so long SRT cues are possible and worth splitting before you use them as subtitles. We did not have a reliable Korean voice sample to test with, so we cannot quote Korean accuracy figures here.
How it works and what stays private
The tool takes the first audio track from your file, converts it to 16 kHz mono and runs a speech recognition model inside your browser. Nothing from the recording is sent to a server. What does come over the network, on the first run only, is the recognition model and its runtime, about 100 MB in total; after that the browser can usually reuse them from its cache. On a slow connection, that first download is the longest part of the job.
Getting a cleaner transcript
- Choose the right language. The tool transcribes in the language you select and does not translate. Korean speech with English selected gives poor results.
- One clear voice works best. Background music, crosstalk and phone-speaker recordings lower accuracy noticeably.
- Trim the silence. Long pauses at the start can shift where segments begin. Cut them with Audio Cutter or Trim Video.
- Check proper nouns. Names of people, places and products are the most common errors, followed by words that sound alike.
From transcript to subtitles
The SRT download writes one numbered cue per segment, with start and end times in hours, minutes, seconds and milliseconds, which is the format most video editors and players accept. Before using it as subtitles, split any segment longer than about seven seconds or two lines, because viewers struggle to read a long block at once. Empty segments are left out of both downloads, and extra blank lines inside a segment are collapsed. If the captions end up slightly early or late against your video, shift them all at once with the Subtitle Converter, or set an offset when you burn them in.
Limits and common problems
Recordings are limited to 60 seconds and 40 MB; split a longer file and transcribe it in pieces. Speakers are not labelled, and punctuation is a best guess. If the page says it could not find a readable audio track, the file has no sound or uses a codec the browser cannot read; convert it with Extract Audio or MP4 to MP3 first. If the speech model could not load, check your connection, since the first run needs to download it. A WAV file is the most reliable input when another format fails.
Frequently asked questions
Does the tool translate Korean into English?
It transcribes in the language you select. Choose Korean for Korean speech and English for English speech.
Is my recording uploaded?
No. The audio is prepared and recognized in your browser. Only the recognition files are downloaded, on the first run.
Can it tell different speakers apart?
No. The transcript is a list of timestamped segments without speaker names. Add names yourself in the text boxes before exporting.
What is the difference between TXT and SRT?
TXT is the plain transcript, one segment per line. SRT keeps the start and end time of every segment, so video players and editors can show it as subtitles.
Do I have to download the recognition files every time?
Usually not. The browser keeps them in its cache after the first run. In our test, transcribing the same 17-second file a second time in the same browser took 11.1 seconds instead of 13.8, because the download was skipped.