A phone, a quiet room, and four minutes in Audacity

The room is the equipment

Amateur recordings sound amateur mostly because of reflections. Sound bounces off hard parallel surfaces, arrives at the microphone a few milliseconds after the direct sound, and the result is the hollow, distant quality everybody recognises as “recorded in a room”.

No processing removes it convincingly. Reverb is far easier to add than to take away, and every de-reverb tool trades artefacts for improvement.

So fix it at the source, for nothing:

• Record in the softest room you have. A bedroom with a bed, curtains and a wardrobe. Not a kitchen, not a bathroom, not an empty office.
• Record close. A hand’s width from the mouth, slightly off-axis so breath does not hit the microphone directly. Proximity means the direct sound dominates the reflections, which is most of the battle.
• Hang something soft behind you if the wall is bare. A duvet over a door, a coat on a hook. This is not a joke; it is what a blanket fort does and it works.
• Turn things off. Fan, air conditioner, fridge, phone notifications. Listen for thirty seconds before you record and you will hear three things you had stopped noticing.

The phone is fine

A modern phone’s microphone is good, and its voice recorder app records at a quality that survives a video edit. Use the native voice memo app rather than recording through the camera, because the camera microphone is optimised for a scene rather than for a voice at close range.

If you have a wired earphone set with an inline microphone, that is often better again, because it sits close to the mouth and away from handling noise.

A USB microphone improves things. It improves them less than moving to a softer room does.

The four-step chain

In Audacity, which is free, open, and runs on machines that will not run anything else. Do these in order.

1. High-pass filter at around 80Hz. Removes rumble, traffic, handling noise and desk thumps that you cannot hear on a laptop speaker and that will be audible on headphones. This single step is the largest improvement available.
2. Noise reduction. Select two seconds of silence, take a noise profile, then apply it to the whole file at a gentle setting. Aggressive noise reduction produces a watery, underwater artefact that is worse than the noise was.
3. Compression. Reduces the gap between loud and quiet, so the voice sits steadily under music and sound effects. Audacity’s compressor with modest settings is enough.
4. Normalise to a consistent peak, around -3dB, so every clip in the project is at the same level.

Four steps, about two minutes once you know where the menus are. Save it as a macro in Audacity and it becomes one click.

Record more than you need

Three things to capture every time, because going back is expensive:

• Thirty seconds of room tone. Silence in the actual room, recorded deliberately. You need this to fill gaps and to smooth edits, and it is the subject of the next lesson.
• Three takes of every line. Storage is free; a second session is not.
• A handclap at the start if you are recording audio separately from a video. It gives you a visible spike and an audible transient to align against, which is what a clapperboard has always been for.

Editing voice

Cut on the breath, not on the word. A cut placed in the middle of a breath sounds natural; a cut immediately after a word ends sounds clipped. Leave a little air.

Remove long pauses and keep short ones. Perfectly even pacing sounds synthetic, which is an interesting reversal: the thing that makes recorded speech sound human is precisely the irregularity that a text-to-speech model struggles to produce.

Do not remove every breath. A narration with no breathing in it is unsettling in a way listeners cannot identify.

Why this beats synthesis for anything long

Stated plainly, because it runs against the direction of the rest of the course: for narration longer than a few sentences, a real voice recorded on a phone in a soft room and cleaned in Audacity is better than almost any synthetic voice you can access for free, and better than several you would pay for.

It is also faster. Ten minutes of recording plus two minutes of processing against an hour of regenerating lines that are nearly right.

Use synthesis for what it is good at (short utterances, many languages, pickups, placeholder tracks while you edit) and record the narration.

Today: record the same paragraph in your kitchen and in your softest room, run both through the four-step chain, and listen on headphones.