Feature Request: Subtitle Post-Processing Module with Professional Formatting Rules

[erklu]

Description

Automatically format subtitles based on professional standards.
The goal is to produce readable and well-structured subtitles following subtitle guidelines such as those from SVT and Medietextarna, including:
• Max 42 characters per line for better readability (we may need to review this for web video; a shorter limit could be more appropriate).
• Maximum two lines per subtitle block.
• Optimized words per minute (WPM) for reading comfort (~160–180 WPM).
• Timestamps aligned with speech for natural flow.
• Export primarily to .srt (optionally also .vtt or other formats).

⸻

Expected Behavior
• Subtitles are divided into blocks based on sentence structure and natural pauses.
• Line length and word count are adapted for optimal readability.
• Display time is adjusted so viewers have enough time to read.
• Speaker changes are clearly marked if multiple people are talking. Ideally, a new speaker starts a new block.
• Exceptions may apply for quick exchanges (e.g., short yes/no answers), where breaking the block could disrupt reading speed.
• Speaker IDs could be optional but not default.
• Noise or incomplete speech is filtered out (e.g., no subtitles for silence or irrelevant sounds).

⸻

Technical Guidelines

Text segmentation into subtitle blocks
• Max 42 characters per line, 2 lines per block.
• Avoid line breaks in the middle of names or fixed expressions.
• Avoid mixing multiple speakers in the same block, unless readability or timing would suffer.

Timing based on reading speed
• Default reading speed: 160–180 words per minute.
• Minimum display time: 1.5 seconds, to avoid flickering subtitles.

Audio-sync and natural timing
• Subtitle should ideally appear 0.5 seconds before speech and remain at least 1 second after the final word.
• If possible, adjust timestamps for natural rhythm in relation to spoken tempo.

Export in professional formats
• Primary export to .srt.
• Support for .vtt or possibly .json if needed for further processing.

[erklu]

A small aesthetic note: In subtitle formatting guidelines from SVT and Medietextarna, it's generally preferred that multi-line blocks are visually balanced. That means if a line break is introduced, line 2 should ideally be slightly longer than line 1 (rather than having e.g. 42 characters on line 1 and only 4 on line 2).

This helps readability and avoids visual "staircases" in the subtitle layout.

Might be worth taking into account when implementing the line wrapping logic.

[erklu]

Status observation – subtitle formatting as of 2025-07-02

Did a review of a recent subtitle export from Transcriber and the following issues have been consistently observed:

Text formatting

• Several lines exceed 60 characters, well above the expected limit of 42 characters per line. Some blocks are over 110-120 characters in total, which will lead to more than two lines, especially on small screens.
• Lines begin with a leading space character, affecting visual alignment.
• From what I can tell, no blocks have more than 2 lines. Good!
• Sentence-final punctuation is inconsistently applied—some completed sentences end without a period.

Timing

• The first block begins at 00:00:00 even though speech starts significantly later, leading to an unusually long display time (~12.5 seconds).
• Some blocks with 15–18 words are displayed for only 4–5 seconds, which results in high WPM (words per minute) rates.
• Subtitle blocks are often placed without consideration for actual speech pauses or natural rhythm in delivery.

These observations reflect the current output as of July 2, 2025, 19:30 CEST.

[erklu]

Done for now. Will need more refinement further on, but ok for first release.

[JohannesGarm]

FIrst version of blocking fix (• Max 42 characters per line for better readability) made with simple --max-len fix. First tests seem promising.

[erklu]

Could improve using #29