sxsphinxstack

Skills / Working / Record clean audio

Working Video & audio skill

Record clean audio

Get clean voice audio from the gear they already own.

Help someone record voice audio that sounds clean, using what they already own — a phone, earbuds with a mic, or a laptop. No purchase is required for a usable result. The session ends with a real recording made, cleaned, and exported, so run a genuine take of whatever they actually need the audio for.

The room beats the mic

Most bad recordings are bad rooms, not bad mics. Before touching settings:

  • Pick the deadest room they have: closet full of clothes, carpeted bedroom, car parked with the engine off. Avoid kitchens, bathrooms, and any room that echoes when they clap once. Do the clap test together.
  • Turn off noise sources: fans, AC, open windows, notifications on every device in the room. Record 10 seconds of "silence" and listen back loud — whatever you hear now will be in every take.
  • Soft stuff absorbs: a duvet over a chair behind the laptop, facing a wardrobe, or recording under a blanket. Judge the setup by the test audio.

Mic and levels

  • Distance: a hand-span from mouth to mic (15–20 cm). Phone mic works fine at this range; the bottom mic of the phone is the good one. Wired earbuds with an inline mic are also solid — clip or hold the mic near the collar.
  • Slightly off-axis: talk past the mic, not straight into it, to dodge plosives (the "p" pops). No pop filter needed at an angle.
  • Set levels with a real test: have them read their actual first lines at performance volume. Peaks should sit around -12 to -6 dB in the meter — loud but never touching the top. On a phone, use a free recorder app that shows a meter (or record a test and check it in Audacity).
  • Record a room-tone clip: 10 seconds of silence in position. The cleanup pass needs it.

Cleanup in Audacity

Audacity is free on every platform. Import the recording, then in this order:

  1. Noise Reduction: select the room-tone section, Get Noise Profile, then apply to the whole track with gentle settings (around 12 dB reduction). Aggressive settings sound underwater — listen and back off if voices go metallic.
  2. Cut the mistakes: false starts, long gaps, mouth noises. Keep natural pauses.
  3. Normalize or Loudness Normalization to about -16 LUFS for spoken word — a standard level that sounds right on phones.
  4. Export WAV for anything they will edit further, MP3 (192 kbps+) for anything going straight out.