How to sync a separate audio recording to your camera footage
Upload every camera and the separate audio file. Vizard Agent cross-correlates the recorder against each camera's own scratch track to find the exact offset, then checks whether the clocks drifted apart across the recording, which is the failure that ruins long dual-system edits.
What is the short version?
Recording sound separately is the right call and it creates one problem: nothing lines up. Finding the offset is easy enough. The part that catches people out is that over half an hour the two devices' clocks quietly disagree.
- Go to Vizard Agent and upload every camera and the audio file.
- Say which audio is the good one.
- Say whether to tighten the edit or keep it complete.
What do you need before you start?
Everything from the shoot, including the camera audio you have no intention of using. Vizard Agent syncs by comparing the recorder's waveform against each camera's own scratch track, so a camera with its microphone switched off is genuinely much harder to place.
- Every camera file. All angles, unedited.
- The audio recording. The one you actually want.
- Camera scratch audio. Bad quality is fine, absent is not.
- The intent. Tighten it, or keep it whole.
- Names for lower thirds. Spelt as they should appear.
What do you type into Vizard Agent?
Say what the pieces are and what the finished thing should be. Vizard Agent works out the sync itself, so the useful instruction is about the edit — how tight, where it should end, and which camera it should favour.
Prompt
Variants worth knowing:
- "Tell me what you cut out." You will want the list.
- Ask for a proxy. A UHD master will not download easily.
- Say how to handle the ending. Cameras stop at different times.
What does Vizard Agent actually do?
Here is the order Vizard Agent worked in on a real thirty-five-minute two-camera interview with separate audio. Steps eight and twelve are the ones that matter: find the offset by cross-correlation, then check whether it stays the same all the way through.
- Checks the format and length of both cameras and the audio.
- Downloads the source footage after checking disk space and speed.
- Extracts the camera audio and sample frames and builds a contact sheet.
- Cross-correlates the audio tracks to find the sync offset.
- Transcribes the master audio and samples more frames.
- Compares the two angles over time and checks for clock drift across the recordings.
- Builds a readable transcript and reads the whole thing.
- Pulls full-resolution frames to make framing decisions from.
- Works out who is on which camera and who is speaking at each moment.
- Renders aligned versions of both cameras after benchmarking the encode speed.
- Gets exact word timings for the introductions and verifies the names.
- Finds the exact word timings at every splice point and reads the wording around each.
- Measures the dialogue loudness and checks the noise floor in the pauses.
- Builds the paper edit and the camera-switch plan, then tests the punch-in framings.
Step six is the one nobody remembers. Two devices started at the same moment still run at slightly different rates, so an edit that is perfectly synced at the top can be a frame or two out by the end — and Vizard Agent checks for that rather than assuming a single offset holds.
What does the result look like?
The master this page is written from measured 3240x2160 at 23.976fps, HEVC, and ran 2094 seconds — about thirty-five minutes of finished episode with the recorder's audio throughout and the picture cutting between the two angles on the speaker.
It came with a 720p proxy, because the full-quality master was over seven gigabytes and would not download reliably. And it came with a list of everything removed, timecoded to the source, after which three specific cuts were put back on request.
When does this not work well?
Dual-system sound has a small number of failure modes and they are all about the recordings themselves rather than the edit. Vizard Agent reports what it cannot align rather than nudging things into place and hoping the lips look close enough.
- No camera audio at all. Nothing to correlate against.
- Recordings that do not overlap. The recorder started late.
- Severe drift. Long takes on cheap clocks need resampling.
- Cameras that stop at different times. Decide what the ending is.
- Names you have not confirmed. It will leave them off rather than guess.
How do you fix a result that came back wrong?
Point at the timecode. Vizard Agent keeps the measured offset, the drift check, the transcript with word timings and the full cut list, so a restored section or a changed camera choice is a re-render from the existing plan.
- "Put that section back." Restored from the cut list.
- "Stay on the other camera there." Switched in the plan.
- "The lips drift near the end." Drift correction applied and re-rendered.
How does Vizard Agent compare to doing it yourself?
By hand this means dragging the audio against the picture until the clap lines up, cutting the whole episode, and then noticing at minute thirty that it has slipped. Vizard Agent measures the offset and then verifies it holds for the length of the recording.
| By hand | Vizard Agent | |
|---|---|---|
| Finding sync | Line up the clap by eye | Cross-correlated from the waveforms |
| Drift | Discovered at the end | Checked across the whole recording |
| Camera choice | Cut by ear | Planned against word timings |
| What was cut | Gone | Listed with source timecodes |
Common questions
Do I need a clapperboard? It helps but is not required. Vizard Agent syncs from the waveforms themselves.
What if a camera had no microphone on? Much harder. Vizard Agent falls back to visible events, which is far less precise.
What is clock drift? Two devices running at very slightly different rates. Vizard Agent checks for it rather than assuming.
Can it handle three cameras? Yes. Vizard Agent aligns each one to the same audio reference.
Will it cut to whoever is speaking? Yes. Vizard Agent works out who sits on which camera and switches on the word timings.
Can I get a 4K master? Yes, and a proxy with it. Vizard Agent knows a seven-gigabyte file is awkward to move around.
What is a proxy for? Scrubbing and further editing. Vizard Agent renders a 720p version alongside the master.
Can it tell me what it removed? Yes, with source timecodes. Vizard Agent produces the full list on request.
Can I put something back? Yes. Vizard Agent restores named cuts from the list and re-renders.
How does it handle the ending? However you say. Vizard Agent finds a natural stopping point before the first camera runs out.
What about the names on screen? It will not guess them. When public sources disagreed on the spellings, Vizard Agent left the lower thirds off and asked rather than putting a misspelt name on someone's interview.
Does it fix the audio levels? Yes. Vizard Agent measures dialogue loudness and the noise floor in the pauses before mixing.
Does this work for music recordings? Yes. Vizard Agent syncs a desk recording to camera footage exactly the same way.