Skip to main content

See how I edit dialogue

A short demonstration for how to make the best out of badly recorded audio. Mainly a high noise floor due to incorrect gain staging during production and other noises due to sub-optimal mic placement.

I will leave out obvious workflow details such as manually adjusting or automating gain before the effects chain as well as de-essing.

Side-chaining the dialogue to a dynamic mid/side EQ that responds by increasing the sides and decreasing the mids, to make room for the dialogue withing the 500-2khz frequencies, the most important area for the human voice. This isn’t only a mix decision, but with more drastic settings also a stylistic one, as the score will appear much wider whenever there is a person talking.

Here is a before and after example. I chose this specific one as it is one where the recording is more than sub-par, recorded with too low gain, so with a very high noise floor.

For natural sounding reverberation, if it is needed at all I like to use a pre-delay time of 20-25ms as this simulates the time that sound takes within a real room to reflect off of the walls. I either use convolution reverb, sometimes using my own impulse responses that I’ve collected. Or if I feel like going digital, chamber reverbs tend to be the best option. Sometimes even just a tiny slapback on 3-4% wet is enough to create that sense of space that I need.

Using very subtle de-noising at the beginning of the chain, in order to prepare the dialogue before the compression stage, to minimise the inevitable amplification of those issues. This is accompanied by simple fixes, like basic EQ’ing such as removing everything sub 80-120hz, removing rumble building up in the mix.

If noise is inevitable, to prevent it from sounding inconsistent I use side chain compression on a noise channel which is chained to the dialogue, which then goes into a gate, cleaning up the lower peaks and creating a sort of inverted envelope follower, which acts as a trigger for me to use on the room tone, attenuating its gain parameter whenever the person in question isn’t speaking.

Sometimes to create a greater sense of space I like to duplicate the dialogue 2 times, hard-panning one of the copies left and the other right and using a simple channel EQ to carve out specific regions with a bandpass, or lowpass. To give this LCR mix some depth, my master gets the mid/side phase treatment (usually not more than 0.2ms). This creates a much greater sense of depth for the vocals if they are hard-panned, especially if the original centred audio channel is sent to an aux channel with a convolution reverb chain as well. This gives the voice back some body.