Sound
Do this in order. Speech first, with nothing else playing. Then the music under it. Then effects. Doing it any other way means redoing the speech once the music arrives, and the speech is the part that matters.
Volume and gain
Select an audio clip and Properties shows two controls that multiply together.
Track volume, labelled with the track's name, applies to the whole lane. This is the right control for a music bed, because the bed is a single thing at a single level.
Clip gain applies to one clip, and it reads out in both percent and decibels. This is the right control for one interview that was recorded quieter than the rest.
Watch the decibel figure rather than the percentage. Lifting a quiet recording brings its background noise up by exactly as much, so there is a point past which you are amplifying the room rather than the voice.
Fades
Fade in and Fade out, in seconds, per clip.
A hard cut through a waveform leaves a step, and a step is an audible click. Two or three frames of fade removes it and is inaudible as a fade. Put a short one at every audio edit. Longer fades, half a second and up, are for joining two pieces of ambience or two sections of music.
EQ
Five bands, with a live equalizer response curve drawn above them so you can see the shape you are making, and an EQ preset menu. Adjust, then name and save your own preset in New EQ preset name.
For a speaking voice, three moves cover most of it.
- Cut the low band hard. Nothing in speech lives below 80 Hz, and traffic, air conditioning, footsteps and table bumps all do.
- A small cut in the low mid, two or three dB, if the voice sounds boxy. That is what a room with parallel hard walls does to a close microphone.
- A gentle lift in the high, about two dB, if the voice sounds dull. Go carefully, because harsh S sounds live in the same place.
Cut before you boost. Boosting raises the noise in that range along with the voice.
Denoise
A toggle per clip. It learns the steady background and subtracts it.
It cannot tell the difference between room noise and the quiet tail of a word, so past a certain point it starts eating the voice and the result goes metallic and underwater. Use it, do not lean on it.
Normalization
Clip normalization per clip, and Final mix normalization in the project properties when nothing is selected.
Clip normalization evens out individual pieces so a quiet interview and a loud one sit at the same level. Final mix normalization sets the whole finished piece to a loudness target.
For web delivery, a spoken word piece at about -16 LUFS integrated with true peak held at -1 works everywhere. Mastering louder than the platform's target gains you nothing, because the platform turns you down and your dynamics are already squashed. See Levels, headroom, and loudness.
Ducking a music bed under a voice
Select the music clip. Properties has Lower [track] under, where you choose which track the music should listen to. Point it at your speech track and the music drops automatically whenever somebody talks.
Five parameters control how it behaves.
| Control | What it decides | Where to start |
|---|---|---|
| Lower [track] under | which track triggers the duck | your speech track |
| Voice sensitivity (dB) | how loud the voice must be to count as speech | set it so breaths do not trigger it |
| Lowering strength | how far the music drops | 9 to 12 dB |
| Lowering speed (ms) | how fast it gets out of the way | fast, so the first word is not buried |
| Return speed (ms) | how fast the music comes back | about 500, so it does not jump up between words |
Less than 9 dB of duck and the voice fights the bed, which is tiring even when every word is audible. More than 12 and the music pumps in and out distractingly. A fast return speed is what makes ducking sound like a machine doing it.
Pan and stereo
Keep narration centre and in mono. It is effectively louder at the same level, it has no phase problems, and nobody has ever complained that narration was not in stereo. Pan ambience and music wide.
Checking it
Use Solo on the track header to hear one thing at a time. Then listen to the whole mix on a phone speaker at half volume, because that is where most of your audience is. If the softest line is unintelligible there, compress harder than feels right on headphones.