Session 3: Audio

There is an asymmetry in video that surprises every beginner: audiences tolerate poor pictures and do not tolerate poor sound. A slightly soft, imperfectly lit video with clear audio is watchable. A beautifully graded video where the voice is distant, hissy or clipping gets closed within seconds, and the viewer cannot explain why.

The reason is that picture is something you look at while sound is something that happens to you. You cannot look away from audio, and distorted or muddy speech forces real effort to decode. When someone says a video feels amateur, they are usually describing the sound, even when they say the camera work.

So audio is not the last thing you polish; it is the thing that decides whether the video is usable at all. The good news is that it is also cheap to fix — the difference between unusable and professional audio is a few decisions about levels, a noise pass and one correct export setting, none of which requires equipment you do not have.

Levels are measured in decibels (dB) on a scale where 0 dB is the maximum and everything else is negative. Going above 0 dB is clipping — hard digital distortion that cannot be repaired afterwards, so the first rule is that nothing ever touches zero.

The targets that work for this kind of video: dialogue peaking around -12 to -6 dB, which is loud and clear with headroom; music sitting 15 to 20 dB below the dialogue, so it is felt rather than fought; and sound effects between the two. If you only remember one figure, remember that speech should peak around -12 dB and never clip.

The related discipline is consistency. A video where one speaker is loud and the next is quiet is exhausting, because the viewer keeps adjusting volume. Normalise your dialogue so every speaker sits at the same level, and check the whole timeline rather than the part you were working on — the quiet section is always the one you did not listen to.

Our footage has the noise problems that Nigerian indoor recordings almost always have: a low hum from the generator or air conditioning, a hiss from the room and the camera preamp, and reverberation from a hard-walled hall with no soft furnishing. Each needs a different response.

Hum is a steady tone at a known frequency and is the easiest to remove — a notch or hum filter at 50 Hz, which is the mains frequency here, takes it out cleanly. Hiss is broadband and needs noise reduction, which works by sampling the noise alone and subtracting it: capture a second of room tone with nobody speaking and use that as the profile. Reverberation cannot be removed after the fact, only reduced slightly, and the real fix is at the recording — a carpeted room, curtains, or a mic closer to the mouth.

Then the limit everyone overruns: noise reduction destroys a voice when pushed too far. Over-processed speech sounds thin, watery and robotic, and it is worse to listen to than the hiss was. Reduce until the noise is not distracting and then stop, and remember the priority — a little hiss with a natural voice beats silence with a robot. Where noise is bad in one section, covering it with B-roll and slightly raised music is often better than processing the voice to death.

Music is chosen by tempo and mood, not by taste. A cut with quick shots wants a track with energy; a reflective moment wants space. The practical test is to play the sequence against two or three tracks and notice which one makes the cutting feel correct — the wrong track makes good cutting feel wrong, which is why music is chosen after the picture is locked rather than before.

The part that costs people money is licensing. Music you like on the radio or on a streaming service is almost never cleared for a client's video, and the consequences are real: Content ID claims that divert a client's ad revenue to the rights holder, muted videos on Instagram and TikTok, and takedowns after the client has already paid you. For a church, a business or any paying client this is your responsibility, not theirs.

So use properly licensed sources: royalty-free libraries with a clear licence, platform libraries where the platform's terms cover the use, or music the client has licensed and given you in writing. Read the licence — some royalty-free tracks forbid commercial use or require attribution. Keep a copy of the licence with the project, because a client asked about their video in two years will come to you, and I found it online is not a defence.

Ducking is music dropping in level while someone speaks and rising again afterwards, and it is the single most effective way to make speech clear over a track. Do it with keyframes — lower the music by about 10 to 15 dB a beat before the speech starts and bring it back a beat after — or with automatic ducking where your editor has it. The common failure is ducking too late, so the first word of a sentence is buried; start the drop before the speech, not with it.

Sound effects work as punctuation and fail as decoration. A subtle whoosh on a transition or a soft click on a graphic can make an edit feel finished; a whoosh on every cut makes it feel like a template. The test is to remove them all and ask whether anything is missing — if nothing is, they were decoration. Silence is also a sound effect, and a beat of total silence before an important line is more powerful than anything you can add.

For voiceover, the equipment matters less than the room. Record in the quietest space you can find — a bedroom with curtains and a bed absorbs far more than a tiled living room — with the mic close, about a hand's width away, and off-axis slightly to avoid breath pops. Hang blankets or record inside a wardrobe if the room is bad. Record more takes than you need, leave a few seconds of silence at the start of each for a noise profile, and never record voiceover in the same room as an air conditioner you cannot switch off.