tella-audio-sync
Audio sync
Some microphones (Bluetooth, some USB interfaces) record the voice a little late, and some cameras deliver their picture later than the microphone does, so the voice lands after or before the lips move. The offset stays the same through a recording, so measure it once per recording and move the clip's whole voice by it.
This skill needs the Tella MCP. If tools such as get_timeline are missing, follow the connection steps in the tella skill first. If update_clip has no audioOffset input, Audio sync isn't available in this workspace: tell the user and stop.
Scope
Fixing sync is a direct action: measure, apply, then report. Ask only when you can't tell which clip is meant. Don't change cuts, layouts or anything else.
audioOffset is how many ms after its picture a clip's voice plays (negative: before), −1000 to 1000. It moves the clip's whole voice, past its cuts, so every cut still removes the words it was made on. get_clip reports it as audioOffset while it is set.
If the user hears it out of sync only in the editor, not in the export or on the share page, that is the editor's playback, not the recording: change nothing.
Measure
Measure with the script in references/measure-lipsync.md, never by looking at frames or filmstrips: a frame is 33 ms and lips are hard to read by eye, so judging tiles gets the direction or the size wrong. The script finds every moment the lips part after being closed (words starting with p, b or m, and the end of closed-mouth pauses), when the voice starts at the same instant, and times the voice against them.