#Translating Subtitles

Subtitle translation is different from book translation. The text must fit timing, reading speed, speaker changes, and formatting tags while still sounding natural. CatScribe's Subtitle Translation screen is a full subtitle workstation: it handles SRT/VTT files, video import with subtitle-track extraction, an in-app player with live subtitle overlay, per-cue timing edits, reading-speed validation, a quality report, and export — including video export with an embedded subtitle track.

Subtitle translation requires the desktop app (it does not run in a plain browser context).

#Supported Inputs

Notes:

• Files must be valid UTF-8. Legacy encodings (Windows-1252, Latin-1) are rejected — re-save the file as UTF-8 first.
• ASS import is not supported. Convert to SRT/VTT before importing.
• Inline <i>, <b>, <u> tags are parsed and preserved; positioning codes such as {\an8} are stripped from the translated text.
• Export uses the project's own format (SRT stays SRT, VTT stays VTT).

#Starting From A Video (No Subtitle File)

You can drop a video directly into "Upload Subtitle or Video":

1. Embedded tracks: if the video contains subtitle tracks, CatScribe extracts them and lets you pick one with "Use This Subtitle".
2. Burned-in subtitles: if no track is found, CatScribe offers "Detect subtitles" — an AI (OCR) pass that reads subtitles rendered into the video image. This requires the optional "AI Subtitle Detection" dependency (~250 MB, installable during onboarding or from Settings → Optional Dependencies) and can take a while on long videos.
3. Sidecar files: if you already have an SRT/VTT for the video, place it next to the video with the same base name and import the video — CatScribe picks it up.

Video processing uses FFmpeg. If it is missing, CatScribe shows "FFmpeg Required" and offers an in-app download (~100 MB).

#Step-by-Step Workflow

1. Open Subtitle Translation and upload the subtitle file (and optionally the video — having the video attached enables playback preview and video export).
2. In "Translation setup" (right panel), either enable Auto Mode and pick a quality level ("Fast" / "Balanced" / "Max Quality"), or configure manually: "Translation API" (engine), source and target languages, and "Context window (entries)" — how many cues are translated together (1–10, default 3). Larger windows give the engine more dialogue context; the setting applies to new uploads.
3. Attach glossary groups if you have them — names and recurring terms drift fast in episodic content.
4. Translate a few chunks first with the per-chunk "Translate" button and review them before committing to "Translate all". You can "Stop" a running batch at any chunk boundary.
5. Use "Lock" on cues you have finalized so batch operations cannot overwrite them, and "Skip" for cues that should keep the original text (songs, on-screen text you do not want translated). Skipped cues export with their original text.
6. Review and edit (see below), then export.

If a chunk fails, CatScribe automatically falls back to Argos when another engine errors, and per-entry retries handle partial responses. A chunk that still cannot complete is left unsaved for retry rather than half-translated.

#Editing And Timing

Select a cue to edit it in the right-hand panel:

• "Original" / "Translation" panes — translation edits save on blur.
• "Edit timing" — adjust "Start" and "End" per cue in MM:SS.d format (100 ms resolution). The editor validates that end is after start.
• The CPS badge shows characters-per-second for the cue, turning amber and then red as reading speed becomes uncomfortable.
• "Retranslate" re-runs a single cue after you adjust settings or glossary.
• The bottom timeline is colored by quality score, so problem regions are visible at a glance; click to jump to a cue.
• With a video attached, the player's "Subtitle display" overlay can show "Translated + original fallback", "Translated only", "Original only", or "Off" — the fastest way to sanity-check timing against actual playback.

An optional "Align" (align-with-audio) action exists for re-syncing cue timings against the video's audio using Whisper, but it is disabled by default and requires both an environment flag (CATSCRIBE_SUBTITLE_WHISPER_ALIGNMENT=1) and a system Whisper/stable-ts installation — treat it as an advanced, opt-in tool.

#Reading Speed And Quality Checks

CatScribe validates subtitles against standard readability limits:

• Maximum 25 characters per second (flagged as too fast above that; the badge warns from 20).
• Maximum 42 characters per line.
• Overlapping cues, negative durations, and empty cues are flagged.

The "Subtitle Quality Report" panel aggregates "Overall", "Translation", "Readability", and "Flow" scores, plus "BLEU Score" and COMET when installed (COMET is an optional ~1–2 GB download offered in-app). Typical flagged issues include "Reading speed is above subtitle comfort range." and "Timing overlap with next subtitle."

"Auto QC Suggestions" proposes mechanical fixes such as splitting an overly long line to improve readability timing.

During review, also check things machines miss:

• Dialogue that sounds unnatural when spoken aloud.
• Split sentences that depend on the previous or next cue — always read neighbors.
• Speaker labels and sound descriptions kept in the target-language convention.

#Keeping Subtitle Tags Safe

If a line contains markup, preserve the tag structure:

<i>I never said that.</i>

Keep the opening and closing tags around the translated text:

<i>Eu nunca disse isso.</i>

CatScribe preserves inline tags through translation for tag-capable engines, but always verify tags on edited lines — a missing closing tag styles everything after it.

#Export

Two export paths:

1. "Export SRT/VTT" — writes the translated subtitle file (default name translated_<original>.<ext>). Skipped cues fall back to original text. If some cues have visual issues, CatScribe asks for confirmation ("Some subtitles may have visual issues. Export anyway?") rather than blocking.
2. "Download video with subtitles" — with a linked video, produces a video file with the translated subtitles embedded as a selectable subtitle track (not burned into the pixels). MP4/MOV output uses an MP4-native text track; MKV/WebM output uses an ASS track, where the "Subtitle export settings" (font size 8–48 pt, font family) apply. This is a Pro feature, unlocked free during the Beta.

#Pre-Delivery Checklist

• Played the video (or a sample) with the translated overlay enabled.
• No red CPS badges on dialogue-critical cues.
• Quality report reviewed; overlaps and negative durations resolved.
• Tags verified on every cue you edited by hand.
• Skipped and locked cues re-checked — skips export original text by design.
• Exported file opens in your target player, and the character encoding renders accents correctly.