Align transcript to SRT timestamps

Each timestamp becomes a numbered cue, while the transcript supplies the exact words viewers will read. Optional reference text tells the aligner how many transcript words belong to each timed segment, so corrections to spelling and punctuation do not destroy the original timing structure. The result uses standard SRT timecodes, stable cue numbering, blank lines between blocks, and a configurable maximum characters-per-line rule.

Prepare the transcript and timing map

Start with the final transcript exactly as it should appear in the subtitles. Whitespace between words is normalized, but spelling, capitalization, and punctuation remain part of the supplied words. Then provide an ordered timestamps array. Every item needs a start and end value in seconds and becomes one SRT cue. A timestamp may also include reference text taken from the original transcription. The aligner does not copy that reference wording into the output; it counts its words and consumes the same number of words from the corrected transcript. This is useful when an automatic transcription already supplied segment times but an editor later fixed names, punctuation, or wording without changing the overall word allocation. When reference text is omitted, the timestamp consumes one transcript word, which suits word-level timing data. Make sure the combined word counts cover the complete transcript. Extra transcript words or timestamp counts that run past the transcript produce an explicit input error instead of silently dropping content. Times use seconds and may include decimals, such as 1.25 or 8.9.

Control subtitle line length

Set max_chars_per_line to the limit required by your delivery workflow, player, broadcaster, or accessibility guidelines. The default is 42 characters, and accepted values range from 8 through 120. Wrapping happens only at word boundaries. The algorithm adds words to a line until the next word would exceed the selected maximum, then starts a new line. It never changes a word, removes punctuation, or inserts a hyphen. Because preserving transcript text is the priority, a single word longer than the requested limit is rejected with a clear error. You can then raise the limit or edit that word intentionally rather than receiving a subtitle that looks valid but contains altered text. Each timestamp remains one cue even if wrapping produces several lines; timing is never invented or divided during formatting. This makes the output predictable when the timing map has already been reviewed. For conventional two-line subtitle targets, prepare timestamp segments with enough words to fit within two wrapped lines at your chosen limit. The capability enforces characters per line, not a hidden cue-length policy.

Validate timing and use the SRT output

Before producing any subtitle blocks, the aligner validates every timing entry. Start and end must be finite, non-negative second values, and the end must be later than the start. Entries must also remain chronological: the next cue cannot begin before the previous cue ends. Adjacent cues may touch exactly, which is common in word-level and tightly edited segment timing. If an overlap or backwards range appears, processing stops and identifies the timestamp item where ordering failed. This fail-fast behavior prevents an apparently polished SRT file from hiding timing defects that players may render inconsistently. Valid seconds are rounded to the nearest millisecond and formatted as the SRT convention HH:MM:SS,mmm. The result includes a complete srt string for copying or saving and a blocks array for applications that want structured cue data. Cue numbers begin at one, arrows and blank separators are inserted automatically, and output is deterministic: identical input always returns identical subtitle text. Run it in the browser for interactive preparation or call the API for $0.002 when subtitle generation belongs in an automated media pipeline.