Captioning Shorts Properly: What Character-Level Timing Means

When captions lag the voice by half a beat, viewers can't name the problem — but your completion rate reports it. The cause: most tools estimate subtitle timing by spreading text evenly at an average reading speed, while real speech speeds up, slows down and pauses.

What character-level timing is

When VoiceSmiths generates a voiceover, the API also returns the start and end time of every character, to the millisecond:

The [0.08s→0.14s]  rain [0.16s→0.34s]  had [0.36s→0.50s] ...

Subtitles stop being estimates — they're derived from the actual audio. Line breaks and disappearance timing follow real speech.

The full workflow

1. Generate: paste your script in the Voice Studio, pick a voice — timing comes included
2. Download SRT: one click in the history panel; cue-splitting by punctuation and duration is automatic
3. Import: CapCut / Premiere / Resolve take the SRT with zero adjustment

Three common variants:

• Have a recording and its script: use forced alignment — more precise than transcription, zero typos
• Only have audio/video: use speech to text — auto language detection with word timestamps
• Multi-language distribution: generate each translated script separately; every language is frame-accurate on its own

One detail

Emotion tags like [excited] shape delivery but are never spoken — and we strip them at export, so subtitles stay clean.

Try a free line on the homepage and hear it yourself.