Voice over technique

How do I fix one bad line in a voice over without re-recording everything?

You punch in at the mistake rather than starting over. Here is what that means, why the run-up matters more than the edit itself, and what has to be true of your recording software for it to work on a single line.

Published by

The short answer

Punch in at the mistake instead of starting again. The recorder plays a few seconds of your own read as a run-up, then switches from playback to recording at the exact point you marked, keeping every word before it and replacing only what follows. This is called punch-back, or punch-in recording. It works cleanly when your takes are bound to script lines, because the software then knows where one line ends and the next begins, and can replace a single line without disturbing its neighbours.

Why re-recording the whole thing is the wrong instinct

The instinct is understandable. You reach the end of a good paragraph, hear yourself fumble one word, and start the paragraph again because that feels like the only clean option. Do it five times in a session and the arithmetic turns ugly: a chapter that takes forty minutes to read takes two hours to capture, and most of the extra time goes on sentences that were already right.

It also costs you the good version. The first pass at a line is usually the most natural one, and by the fourth attempt you are performing the sentence rather than reading it. Keeping what was already good is not only faster, it is usually better.

What punch-in recording actually is

Punching in means dropping into record part-way through an existing take, so only the bad section is replaced. It comes from multitrack music recording, where re-playing an entire song to fix one note was never practical. Three things happen in order:

  1. You mark the punch point. The moment where the good part stops and the replacement should begin.
  2. The recorder pre-rolls. It plays back the seconds immediately before that point so you can hear your own pace, pitch and energy.
  3. Recording takes over at the mark. Everything before it is kept exactly as it was. Everything after it is replaced by what you say now.

The pre-roll is the part people skip, and it is the part that decides whether the edit is audible. Coming in cold almost always lands on a slightly different pitch or pace, and the ear is very good at hearing that seam. Performing along with a few seconds of your own voice puts you back in the same read.

Why this is hard in a normal audio editor

Every DAW can punch in. What a DAW cannot do is know what you were reading. Your recording is one long waveform, and the boundary between the line you flubbed and the line before it exists only in your head and in the script beside you. So the work becomes manual: find the passage by eye, scrub to confirm, set an in point, set an out point, record the replacement, nudge it into place, cross-fade both joins.

That is a perfectly good method and plenty of narrators work that way for a living. It is just that all of it is bookkeeping the software could have done, because the information it needs, which line is which, was available at the moment you recorded and then thrown away.

Four ways to fix one line, and what each costs

  • Re-read the whole take. The most time, and usually the loss of the most natural version of every line you already had right.
  • Manual punch in a DAW. Finding both boundaries by eye and ear, every time, then cross-fading two joins.
  • Record each line as its own file. Almost nothing, if the software does it for you. A retake is just that line again.
  • Punch in against the script. Almost nothing, if the software tracked which line you were on while you read.

The thing that makes single-line repair possible

A take that is bound to a script line. If the software recorded line 12 as line 12, rather than as audio between 4:31 and 4:48, then replacing line 12 needs no measuring. You point at the line, you speak it again, and the new audio lands where the old one was. Everything after it keeps its own audio and shifts to make room if the new read runs longer.

That is the design premise of Medrak Recorder: the script is not a document you read from beside the recorder, it is the structure the recording is stored against. Each spoken line keeps its own take, and a long continuous read still carries markers for where every line began, so either way the boundaries are known rather than hunted for.

How to do it, step by step

  1. Keep reading. When you flub a word, do not stop and restart from the top. Stop where you are.
  2. Put the cursor at the mistake. Click the waveform at the word that went wrong, or simply press stop, which leaves the cursor exactly where you stopped.
  3. Choose the run-up. From the start of the previous line, from two lines back, or a fixed five, ten, twenty or thirty seconds.
  4. Punch. Press P. You hear your own read approaching the punch point, a marker passes, and the microphone takes over. Read on.
  5. Stop when the line is right. Everything before the marker was never touched, and only what you replaced has changed.

The one-press variants are worth learning: 1 punches with the previous line as run-up, 2 with two lines, and 5 with five seconds, without touching a dropdown.

Making the repair inaudible

The splice itself is sample-accurate and adds no artefact. What gives a repair away is physical, so the fixes are physical too.

  • Fix it in the same session. A voice at 9am and the same voice at 6pm are different instruments.
  • Do not move. Distance and angle to the microphone change tone more than most people expect. Mark your position if you have to step away.
  • Perform the run-up, do not just listen to it. Read along silently so you arrive at the punch point already in the take.
  • Punch at a natural gap. The start of a word, after a breath, is far more forgiving than the middle of a held vowel.

What this does not solve

Punching in replaces a performance, not an environment. If the room tone changed because a fridge started up, or a window was open for the first pass and shut for the second, the join will be audible however good the edit is, and the honest fix is re-recording the affected stretch under conditions you can reproduce. It will not rescue a take whose problem is pacing across a whole paragraph either, because by then the thing that is wrong is the paragraph.

Related reading