Karaoke subtitles: word-level sync explained
Subtitles that light up word by word are now the norm in short-form video. Their quality depends entirely on something invisible: the timing precision of each individual word.
Transcription and alignment are two different problems
Automatic speech recognition answers "what was said?". Alignment answers "at exactly what moment was each word spoken?". These are different tasks, and most tools only handle the first.
The symptom is recognisable: subtitles are broadly correct but the animation lags by half a second, or a word lights up just after it was spoken. Across a thirty-second Reel, it reads as a manufacturing defect.
The rewritten-text problem
A second, subtler flaw appears when the video starts from a written script. Automatic transcription never returns the script exactly: it normalises numbers, corrects hesitations, drops a proper noun, changes punctuation.
The result is that the subtitles on screen are not your text but the model’s interpretation of it. For a brand that is a real problem — a misspelled product name on screen costs more than a timing offset.
The fix is forced alignment: keep the exact wording of the original script and use speech recognition only to recover timings. The displayed text stays what you wrote, with timing derived from the actual audio. This is the approach Viffly uses for subtitling script-driven productions.
Style matters as much as timing
Once timing is right, the result comes down to style. The parameters that genuinely change mobile legibility:
- Background contrast. A transparent subtitle with a drop shadow works over a dark image and becomes unreadable over a bright one.
- Size. Vertical format tolerates far larger type than 16:9 — that is what makes "highlight" styles effective.
- Words per line. Past four or five, the eye disengages before the end of the line.
- The accent colour, which must stay identical across the whole video to read as a system rather than an effect.
These choices are exactly what templates codify: each one fixes a coherent combination of background, size, typography and animation.