How to cut around the moments a guest's microphone dropped out
Tell Vizard Agent that a guest's microphone cut out during the recording. It searches the transcript for the moments you mention, scans the rest of the show for silent gaps that look like the same fault, reads word-level timings through each choppy stretch, and builds the cut plan so every removal lands between words.
What do you type into Vizard Agent?
Ask the second question as well as the first. "Clean up these three moments" gets you three moments cleaned; "find any others" is what turns a patch into a pass, and on a long interview it almost always finds something you had not heard. If you're unsure of the times, say "a few times" rather than guessing, so Vizard Agent's search isn't anchored to the wrong place.
Prompt
What does Vizard Agent actually do?
A dropped microphone doesn't sound like silence in the room, so hosts catch two or three and miss the rest. The transcript catches all of them, because a dropout leaves a hole in the words rather than a pause between them. A pause has room tone in it and a dropout does not. In the session behind this article it:
- Searched the transcript of a podcast episode for the moments the host described, and read around each one to see how much speech was affected.
- Scanned the whole show for silent gaps that looked like the same fault, then downloaded the source to examine a suspicious stretch closely.
- Read the word-level timings through the choppy stretch, then re-read it to separate lost words from ordinary pauses.
- Built the cut plan from the word timings, test-rendered a section and checked it on frames and waveform, then had the cut sections watched back for speaker accuracy and smoothness.
- Rebuilt the episode around the shortened answers and re-checked the open and the close before uploading.
When does this not work well?
Cutting only works when there is something on either side worth keeping. If the microphone was down for most of an answer, or the key sentence is the part that vanished, no amount of clean cutting by Vizard Agent brings it back. One microphone for everyone, heavy background music or overlapping speech also defeat it.
Common questions
Do I need to know when it happened? No. Vizard Agent scans the whole recording either way.
Will the guest sound choppy? Not if the cuts land between words, which is what Vizard Agent plans for.
Why look at words rather than the waveform? Because a dropout and a thoughtful pause look almost identical in a waveform and completely different in a word-timed transcript.