Turn a recording into text you have checked
A video transcript is the spoken words from a recording, with available cue times for review. This video to text tool processes local files in the browser, supports English speech, and accepts files up to 100 MB when the browser can decode their audio.
1. Which video file should I open first?
Choose a local MP4, MOV or WebM file with clear English speech and a size below 100 MB. Drop it on the upload area or choose browse. Check the selected file, then choose Transcribe this file. This page accepts local audio too; the public practice WAV lets you try the workflow without a private recording.
The container extension does not guarantee that the browser can decode its audio. A screen recording must actually contain the voices you need. This page has no video-link field, audio-track selector or format converter. For unusual containers, consult the Recording Format Guide.
Check: The intended local file is selected and its audio includes the speech you need.
Back to the tool2. Why does the first transcription take longer?
The workflow decodes audio, checks its language, loads the English recognition model and transcribes speech. The first use downloads both the language detector and English model, so keep a working connection and allow the stage messages to change. Model loading and recognition have no fixed completion time.
Cached model resources can help later runs, but the English instance is released before a new language check. A cache hit does not mean all initialization disappears. The language check samples the beginning of the recording. Clearly detected non-English speech, uncertain language and technical detection failure have separate notices and stop without producing a transcript.
Check: Read the current stage; a failed language check is not evidence that the speech is non-English.
Back to the tool3. How do I check and correct the transcript?
After transcription, use the local recording player and click a cue timestamp to replay from its start. Listen for names, numbers and missing words, then edit the corresponding text field. Copy and downloads use the latest wording. The “[reviewed]” marker in the illustration was added solely to demonstrate that an edit reaches the saved file.
The recognition result is a draft, especially with music, overlapping voices or unclear speech. Playback depends on browser codec support. Wording changes do not retime cues. When the cues have complete valid timing, Open in Subtitle Editor sends the transcript and recording directly to another same-origin tab for timing edits. If the tab is blocked or transfer fails, download SRT and open it there manually.
Check: Compare the actual words with the recording, then verify the edited text.
Back to the tool4. How do I find a word in a long transcript?
Use Find in this transcript after recognition. Previous and Next move through matches in the latest edited words. Choose Edit this match to select its text, or play its cue when a recording and start time are available. Enter moves forward, Shift+Enter moves back, and Escape in the search box clears the query.
Matching is literal, ignores ASCII letter case and stays within one cue. A phrase split between cues will not match across that boundary. Search does not alter words, timing or exports, and the query is not uploaded or saved.
Check: Search a distinctive word within a cue and review the match in context.
Back to the tool5. Should I save TXT, Markdown, SRT or VTT?
Copy and TXT use edited words, with no times by default. Enable Include timestamps in TXT, Markdown and Copy to add available start times. Markdown saves a generic title and one paragraph per cue, escaping Markdown punctuation and HTML-like words. It contains transcript text, not an AI summary, and omits your original filename.
SRT, VTT and editor transfer require non-empty text and complete valid start/end times. Missing ends are not estimated. If those controls are unavailable, read the export message and save TXT or Markdown instead. Download before closing the tab; there is no automatic cloud save or transcript history. A subtitle file does not burn captions into your video.
Check: Reopen the download and confirm your latest wording, format and desired timestamps.
Back to the tool6. How do I recover from an error or cancel a run?
For a decode error, check that the recording has a supported audio codec; try a compatible export such as MP4 with AAC, or a playable WAV/MP3. Renaming a file extension does not convert it. The error illustration was produced using an intentionally invalid audio file. A technical language-check failure instead calls for checking the connection and retrying, not assuming the audio is non-English.
Cancel processing discards the task’s result. A model download or local audio/model operation already in progress may need to finish before another task can start. After failure or cancellation, Try the same input again reuses the selected file in this tab. Selecting a different input replaces that retry target; only one task runs at a time.
Check: Wait until controls unlock, then retry the retained file or select a compatible recording.
Back to the toolMore questions and common problems
Can this transcribe speech in another language?
Speech recognition is English only. Non-English, uncertain language and detection failure stop separately. Editing subtitle text in Subtitle Editor has no English-only restriction because you supply the words.
Can I paste a YouTube or Instagram URL here?
This page uses local files. For public Instagram media use the Instagram Transcript tool; its URL lookup involves servers. For existing YouTube captions use YouTube Transcript Download. A caption lookup is different from speech recognition.
Does my recording or transcript leave this browser?
Local file decoding, recognition, edits, playback and exports happen here. Model files are downloaded from delivery services; your local recording is not uploaded. Editor transfer uses same-origin browser messaging and does not put transcript text in the URL or save a cloud project. See the privacy policy.
Can I use this on a phone?
You can select files in a mobile browser, but decoding support, available memory and processing time vary. A small clear recording is a useful first check.
Historical measurement notes
These are previously published observations, not new measurements for this walkthrough or a prediction for your next run.
Read the historical method, results and limitations
These are historical measurements from 2026-09-14 using only whisper-tiny.en. They do not measure the current language-detection plus English-transcription workflow, and exclude Instagram lookup and media download.
| Audio length | Historical wall-clock time | Historical model load |
|---|---|---|
| 11 s | 16.89 s | included |
| 11 s | 6.49 s | already in memory |
| 30 s | 8.98 s | already in memory |
| 60 s | 16.06 s | already in memory |
| 120 s | 28.58 s | already in memory |
In the historical test, the first run included model loading; subsequent runs retained the English model in memory. Current jobs release that instance before language detection, so a resource cache hit still requires initialization. Wall-clock time is not CPU time, and subtracting two runs does not isolate download or initialization time.
One machine, one WASM browser, one run per historical data point. Audio clips were constructed from a repeated public JFK recording. These rows do not predict today's end-to-end time, other devices or other recording content.
Sources: historical raw CSV · current two-model measurements (2026-10-14) · method, environment and limitations.