What actually gets transcribed
Konthora transcribes the audio track of a video, not the picture. That distinction matters for two reasons. First, a video with a silent or music-only audio track produces no text, however interesting the visuals are. Second, burned-in text in the video frame is not read, because it is not audio.
If your video has narration, it transcribes that. If your video is a screencast with no voiceover, the transcript will be empty, and no transcription tool can help — the information is in the frames, not the track.
Upload the original rather than a heavily compressed re-encode where you have a choice. Video platforms routinely transcode to low bitrates, and the artefacts they introduce are audible to the model as garbled speech.
Why word-level timing matters for captions
Captions are read at roughly 160 to 180 words per minute, and a reader needs the text to arrive when the words are spoken. Sentence-level timing puts a whole sentence on screen for its entire duration, so the viewer reads a line while the speaker has already moved on.
Word-level timestamps fix this: the cue reflects what is actually being said at that moment, so the captions track the voice instead of lagging behind it. This is the single biggest quality difference between an automated caption file and a usable one.
There is a cost. Word mode produces more cues and a larger file, and some older players handle dense cues poorly. If your audience is on a modern HTML5 player, take the improvement.
Caption standards and reading speed
Two lines maximum per caption is the near-universal convention, and going beyond it means the viewer is reading rather than following. Long sentences need splitting into more cues, not wider ones.
Characters per line matter for the same reason. Roughly 32 to 42 characters per line is the working range; beyond that, the eye loses the start of the next line while still finishing the current one.
Sound that is not speech has no correct caption. Laughter, applause and music are normally left uncaptioned, and a file that invents text for them is worse than a file that stays quiet.
Captions as a search asset
On YouTube and most platforms, caption text is indexed alongside title, description and tags. That makes a caption file a second and third chance at the same query, and for a video whose title is vague it is often the only text that matches what a viewer would actually search.
It also carries the spoken words. If your video says something the title does not, a transcript-only competitor can outrank you on it — the reverse of the usual situation, and worth checking for.
Publishing the transcript on your own site as well is worth doing for the same reason as podcasts: you want the indexed text on a domain you control, not only inside someone else’s platform.
Step by step
Upload the video file directly
MP4, WebM and MOV are all accepted. Konthora extracts the audio track automatically and transcribes the speech in it, so you do not need to pull an audio file out first.
Choose word-level timestamps for subtitles
Subtitle quality depends on tight timing. Word mode gives each word its own timestamp, which is what keeps captions in sync with the speaker instead of trailing them.
Run the transcription
Processing is CPU-bound and takes roughly the length of the audio. Very long videos need splitting first, since the limit is 10 minutes per file.
Proof the captions against the video once
Watch the first two minutes with the captions open. Non-speech sound — laughter, applause, a door — has no correct caption, and a caption file that desynchronises early puts a viewer straight out of sync for the rest.
Upload SRT or VTT to your platform
SRT is the near-universal subtitle format; VTT is what HTML5 players want. Keep the file after uploading, because platforms re-generate captions and yours is the version with the timing you intended. Download it within the 60-minute retention window, since a caption file is much more work to recreate than an upload is to retry.