Music videos, lyrics and the limits of speech recognition
Singing is not speech: pitch, sustain and mixing break the models that handle conversation well. A music video will transcribe, but treat the text as a draft.
Translation happens after the timing is fixed, which is why the subtitles stay in sync.
before우리 내일 오전 열 시에 공항으로 출발해요.
after · english subtitleWe leave for the airport at ten tomorrow morning.
00:00:12,480 → 00:00:15,120 · one cue, one line
Run one clip free
Free under 60 seconds, no account. Longer videos: 1 credit per minute.
What happens to your video
- Upload the file or paste a public link. Phone footage (HEVC/MOV) is read as it comes off the phone.
- The audio is transcribed first, with word-level timing, so every cue starts where speech starts.
- Each line is translated on its own. Timecodes are copied through and never rewritten.
- Download the SRT, or a copy of the video with the subtitles burned in at the original resolution.
Notes on the audio
- Clear vocals over a sparse mix work best; dense production does not.
- The timing will follow the vocal line, not the beat.
- For lyrics that must be exact, send an existing text file instead.
What it costs
Free under 60sno signup, no watermark
1 credit / minute$1 buys 20 credits
Both outputsSRT, or burned-in video
Questions
Can it translate a song accurately?
Not reliably. Singing plus music is outside what speech models handle well.
What if I already have the lyrics?
Then translate the text file; that path is exact.
Related