packages/docs/docs/recorder/editing/silence-removal.mdx
When captions are generated, the Whisper.cpp timestamps are taken to determine the start and end of the transcription.
Therefore, you should make sure captions are generated before you manually trim the video.
The easiest way to fix mistakes is to click on a caption and edit the transcript.
The first non-empty token determines the start trim.
The last non-empty token determines the end trim.
In the right hand side default props editor, apply a startOffset to the scene.
The offset is in frames.
A negative offset will start the video earlier, a positive offset will start the video later.
To change the end trim, apply a endOffset to the scene. It follows the same logic as the start offset.
The logic for determining the start and end timestamp is in remotion/calculate-metadata/add-metadata-to-scene.ts.
The default start padding is a 0.25 seconds and the default end padding is a 0.5 seconds.
These values are defined in remotion/calculate-metadata/get-start-end-frame.ts