Captions track exceeds edited video length, causing captions to cut off prematurely

[zebisko]

Description

After editing a video and clicking Generate Captions, the captions track is generated based on the original (pre-edit) video length rather than the edited length. This causes the captions track to be longer than the actual video track, which results in captions stopping prematurely mid-video. There is no way to scroll beyond the visible timeline or add/edit captions in the overflow region.

Reproduction

1. Record a video in Cap
2. Edit the video (trim, cut, etc.)
3. Click Generate Captions on the edited video
4. Observe that the captions track extends beyond the video track
5. Captions stop prematurely — the overflow portion of the captions track is inaccessible

Additional Context

• Cap version: 0.4.81 (0.4.81)
• Operating system, version: Tahoe (26.3.1)

• Device (optional): Apple M1 Max

[linear]

CAP-667

[tembo]

This is a known issue that was previously reported in #1522 - the caption generation transcribes the original raw audio files from the recording, not the edited/trimmed version.

Looking at the code flow:

1. When you click "Generate Captions", it calls transcribeEditorCaptions(editorInstance.path, ...) in apps/desktop/src/routes/editor/CaptionsTab.tsx:349-352

2. The Rust transcribe_audio command in apps/desktop/src-tauri/src/captions.rs extracts audio from the .cap project by reading recording-meta.json and concatenating the raw audio segments (lines 62-489)

3. The audio extraction reads directly from segment["mic"]["path"] and segment["system_audio"]["path"] without applying any timeline cuts/trims that exist in the project's timeline.segments

The fix would require either:

• Option A: Export a temporary edited audio file that respects the timeline cuts, then transcribe that
• Option B: Post-process the caption segments to filter/adjust timestamps based on the timeline cuts

The totalDuration() function in the editor context already accounts for edits (it calculates the edited timeline length), but the caption generation doesn't use this - it transcribes the full raw recording duration.

For a workaround until this is fixed: you can manually adjust caption timings using the caption track in the timeline (drag handles to resize, or use the start/end time inputs in the Captions tab when a segment is selected).

[richiemcilroy]

Thanks for reporting! Will be fixed later today in 0.4.82