Editing Speech Without Making It Robotic

A good speech edit is invisible. The listener never thinks "there was a cut." A bad one jumps out, usually as a click or a breath chopped off halfway. Over-edit a whole interview and the rhythm gets so tight it sounds like a machine reading a list.

Before any of this, one rule for journalists. Editing for clarity is fine: ums, false starts, a cough, a phone ringing. Editing that changes what someone meant is not. Never splice two answers together so a person seems to say something they did not, and never cut a qualifier ("I don't think the board should...") to flip the meaning. If you would be embarrassed to play the raw tape next to the edit, don't make the edit.

Where to cut

Speech has natural seams. Cut at one and the join disappears.

• In a pause or breath. The best seam. Cut in the gap, and the voice on both sides starts and stops cleanly.
• Just before a stop consonant. P, B, T, D, K and G all start with a split second of silence while the mouth closes. Cut right there, at the left edge of the sharp burst you see on the waveform, and the consonant's own attack covers the join.
• Not in the middle of a vowel, an M, an N or an S. These are continuous sounds. Chop one and the ear hears a jump, even with a fade.

Zoom in to find these. In SoundStudio press + to zoom, or use Ctrl+mouse wheel around the pointer. A stop consonant shows as a sharp spike after a flat spot. A vowel is a fat, regular wave. With practice you will read words off the waveform.

Making the cut

Drag across the part you want gone and press Delete (or Ctrl+X to cut it to the clipboard). Play from a second or two before the join with Space. Listen at least twice.

Every cut joins two points of the waveform. If one side is high and the other low, the wave jumps, and an instant jump is a click. The theory page explains it at Clean edits. SoundStudio handles the basic case for you: Cut, Delete, Paste, Duplicate and Remove Silence blend the join over 6 ms, and Crop to Selection and Trim Start/End add a 3 ms fade at the new edges.

For bigger joins, where two different takes meet, use a longer crossfade by hand. SoundStudio has no single crossfade button; the Tools tab shows a tip instead: fade out the end of the first region and fade in the start of the second. Select the last 20 to 50 ms before the join, press Fade Out; select the first 20 to 50 ms after it, press Fade In. Short fades on speech. Long fades on speech sound like someone turned a knob.

Ums, false starts and the robot problem

The temptation is to take out every um, every uh, every "you know." Do that and the speaker starts to sound like a hostage reading a statement. People need some hesitation to sound like people.

A workable rule: cut the fillers that distract, leave the ones that carry thought. An um before a hard answer tells the listener the person is thinking. Four ums in one sentence tell the listener nothing and make them wince.

• Single filler in a pause: cut the um and most of the pause around it, but leave a little air. Taking out the filler and all the silence pulls the words too close.
• Filler glued to a word ("andum"): often leave it. Cutting it usually chops the word.
• False start: keep one complete attempt. Cut from just before the first try to just before the second, ideally at a stop consonant.
• Long pauses: shorten them, don't delete them. A 2-second pause can become 0.6 seconds. A pause of zero sounds like a skip.

Room tone fill

When you delete a cough or a door slam in the middle of a pause, you may want to keep the pause length. Deleting it closes the gap. Instead, copy a bit of your recorded room tone (Ctrl+C), delete the noise, put the playhead at the join and paste (Ctrl+V). In SoundStudio, Paste inserts at the playhead; it does not replace a selection. The background stays continuous, so the listener never hears a hole.

Never fill a gap with digital silence on a voice track. Dead air under room noise is easy to hear, like a radio cutting out.

Breaths

Breaths are normal. Big gasps right in the mic, after compression has pulled them up, are not pleasant. The fix is usually to turn them down, not out.

SoundStudio's Voice tab has a Breath Remover. It looks for noise with no pitch, quieter than the words, at least a tenth of a second long, and fades the edges over 12 ms. Your options:

• Action: Turn down by 3 to 40 dB, Silence, or Cut out. The help page suggests 15 dB as a good start. Cut out also shortens the file, which is handy for a timed spot and too tight for most interviews.
• Sensitivity: Gentle, Normal or Strong. Strong can catch soft F, TH and H sounds, so review it.
• Find and Review lists the breaths first. Click a row to hear it, untick any that are not breaths, then press Fix Ticked Breaths. Do this the first time you work with a new voice or room. Find and Fix All Breaths does everything in one click, and one Ctrl+Z undoes it all.

Tidy the ends

Trim Start/End, under Silence Removal in the Noise tab, removes silence at the head and tail. Leave about a quarter to half a second of room tone before the first word and after the last, so the voice does not start abruptly. Remove Silence, in the same place, cuts every quiet stretch below a Threshold that lasts longer than Min Duration. It is fast and blunt. On a conversation it can make the whole thing feel breathless. Use it on rough drafts, not finished interviews.