Automatic subtitles
Generate subtitles automatically
Automatic subtitling is speech recognition plus line-fitting plus rendering. Knowing which of those three is failing is most of what makes the difference between a usable result and a frustrating one.
What actually happens to your video
Audio is separated from the picture
The soundtrack is extracted, because the transcription step only needs the audio.
Speech is transcribed with timings
A speech recognition model produces the words and the moment each was spoken.
Text is translated, if you asked for another language
The transcript is translated with surrounding context rather than line by line.
Lines are fitted and rendered
Text is split into readable lines within the timings available, drawn into the frames, and encoded back to MP4.
What makes automatic subtitles good or bad
Almost every quality complaint traces back to the audio rather than the model. These are the factors that matter most, roughly in order:
- Speech clearly above the background. Music mixed at the same level as dialogue is the most common cause of a poor transcript.
- One person at a time. Overlapping speech is genuinely ambiguous, and no amount of model quality resolves it.
- Vocabulary the model has seen. Common words transcribe reliably; invented product names and internal acronyms do not.
- Consistent recording conditions. A clip that switches between a close microphone and phone audio produces uneven results across the same file.
What to check before you publish
- Proper nouns, in the first pass. They carry the most meaning and have the least context to recover from.
- Numbers, especially in a pricing or measurement context.
- The first and last thirty seconds, where openings and sign-offs often get clipped.
- Any stretch where you know the audio was difficult. If it sounded rough to you, it was worse for the model.
Automatic does not have to mean final
On the Pro and Studio plans the generated subtitles are a starting point rather than a finished product. You can edit wording and timings in the subtitle editor, trim the clip, and reburn. That combination, automatic first pass plus a human correcting the few lines that matter, is what makes subtitling fast without making it sloppy.
Generate subtitles for your first video with 50 free credits.
Continue with Google50 free credits. No credit card needed.Frequently asked questions
Do I need to type anything?
No. The subtitles are generated from the audio. You can edit them afterwards on the Pro and Studio plans if you want to correct a line.
How accurate are automatic subtitles?
It depends almost entirely on the audio. Clear speech with little background noise transcribes well; overlapping speakers and loud music do not. Reviewing proper nouns and numbers catches most of what slips through.
Does it work with accents?
Generally yes, for widely spoken accents. Very strong regional accents and heavy background noise are the hardest combination, and that is where reviewing the transcript pays off most.
Can it handle more than one speaker?
It transcribes what is said regardless of who says it. It does not label speakers, and sections where people talk over each other are the least reliable part of any transcript.
What does it cost to try?
Fifty free credits, which is about ten minutes of video, with no card required.