Good Audio for Screen Recordings on a Budget: Mic, Room, Levels

Budget tutorial audio tends to fail in the same few ways. The voice sounds like it was recorded in a bathroom, the p's thump, the loudest sentence clips, a fan hisses underneath everything, or the video is so quiet that viewers turn their speakers up and get blasted by the next thing they play. Each has a cause you can find in five minutes, and most cost little or nothing to fix.

This guide gives you an order of operations instead of a shopping list. You will choose a mic for your room, position it, set the input level with numbers rather than guesses, apply a light cleanup chain, export at a sensible loudness, and run a five-minute test recording that shows which problem to fix first.

Fix audio in this order: why the sequence matters more than the gear

Each stage can only improve what the one before it delivered. Work down this ladder and spend money on the first rung that fails.

The stakes are more than taste. In two experiments, Eryn Newman and Norbert Schwarz ("Good Sound, Good Research", Science Communication, 2018) played identical conference talks and NPR Science Friday interviews in high or low audio quality. Per the abstract, listeners rated the research and the researcher less favourably when the audio was poor. The participants judged scientists, not software walkthroughs, and the abstract gives no sample sizes or effect sizes, so read it as a direction, not a measured size.

Microphone, placement, and room: the three choices that fix most problems

Choose the type for your room

Shure's educational guide to recording describes the trade-off. A dynamic mic is rugged and handles loud sound without distortion. A condenser needs phantom power (a small voltage an interface or mixer sends down the cable) or a battery, and is more sensitive, with some self-noise, more in cheaper models, that shows as hiss at high gain. USB condensers draw power from the USB port instead. In a poor room, Shure says to put the mic very close to the sound source.

So: echoey or noisy room, a dynamic mic used close; quiet, soft room, a condenser is fine; no time to learn technique, a headset.

For a first mic we would look for four things, none of which need a brand name: cardioid pickup (omnidirectional and stereo settings pick up the room as readily as your voice), a mute button or physical gain control, a headphone jack so you hear yourself without delay, and a stand or arm that is not on the desk you type on.

Place it a hand-span away, off-axis

Shure's spoken-word advice applies directly to screen recording:

• Mic 6–12 inches (15–30 cm) from your mouth. Closer, and proximity effect (the bass boost directional mics add up close) makes the voice muddy.
• Aim from slightly above or below your mouth, not straight at it, to soften the air burst on p and t.
• Pop filter in front; it also shows you how far to sit.
• Nothing hard under the mic. Tabletops reflect sound back into it.
• Face the mic when you speak. High frequencies are directional, so a turned head dulls the voice.

Distance sets the voice-to-room ratio that no later step can improve. Per Shure, each doubling of distance drops the direct sound by 6 dB, while ambient room sound stays at nearly the same level. Moving from 6 to 24 inches, two doublings, costs about 12 dB of voice, and winning it back with gain lifts the room, the fan and the preamp hiss by the same amount. The rule is an idealisation: past what Shure calls the critical distance the room dominates, and Sonarworks puts that at 30–60 cm in typical home rooms.

Mount it off the desk and between you and the screen

Screen recording adds problems a podcast does not have.

• Keep the mic off your typing surface. Keys and clicks travel through a desk, and a stand on it can carry them to the mic. That is physical reasoning, not a cited figure, so test it by tapping the desk during a take. A boom arm or a stand on a separate surface breaks the path.
• Put the mic between your mouth and the display, aimed at your mouth, so a glance at the screen does not turn you off-axis.
• Choose how you narrate. Narrate live for short, conversational demos. When you type a lot or stumble, record the screen first and add the voice-over afterwards: no keyboard under your voice and a retake per sentence, at the price of rehearsing the timing. Our guide to recording software tutorials people finish covers per-step retakes.

Treat the room in ten minutes

Work down this list and stop when the clap test (item 3) sounds dry.

1. Remove noise at the source. Switch off spare fans, close the door and window, and silence notification sounds. Our notification audit covers Mac, Chrome and phone.
2. Soften what your voice bounces off. Sonarworks says the wall directly behind the mic typically creates the most problematic reflection. Hang a blanket or coats there, and put a towel under the mic stand. A cardioid mic is least sensitive directly behind it, at 180 degrees, per Shure, so point that side at the computer or window.
3. Clap once and listen. Sonarworks suggests this test: a good room sounds controlled, without obvious reflections. Rings or slap-back mean hard surfaces are returning your voice (our reading).
4. Face an open closet rather than recording inside it. Sonarworks says a closet sounds boxy.

Set the input level by the numbers, not by feel

dBFS means decibels relative to full scale. Zero is the digital ceiling, and a signal that tries to exceed it is clipped into harsh distortion. How far below zero should your loudest word sit?

Advice differs by a few decibels. A 2010 Sound On Sound answer about music puts peaks around -10 dBFS, Sonarworks' gain staging guide calls -18 to -12 dBFS the usual recording level, and its home vocal guide says to keep vocal peaks between -12 and -6 dBFS. For speech, aim your loudest words at -12 to -6 dBFS.

1. Sit where you will record, at your final mic distance.
2. Open the input setting. On macOS 13 Ventura or later: Apple menu > System Settings > Sound > Input, select the mic and drag the input volume slider (Apple's guide; earlier versions use System Preferences). On Windows 11: Start > Settings > System > Sound, select the mic under Input and adjust Input volume (Microsoft's guide, which also covers Windows 10). With no software slider, use the mic's gain knob, and adjust only one of the two.
3. Record about 15 seconds of the loudest part of your script, at the energy you will really use.
4. Measure the peak with the command below and read max_volume. FFmpeg is free; on a Mac with Homebrew, brew install ffmpeg. Without a terminal, Audacity's recording meter has a dB scale with 0 dB as the clipping point and turns yellow near -6 dB (manual), so keep peaks out of the yellow.
5. Adjust and repeat. The sliders carry no dBFS numbers, so expect two or three passes.

ffmpeg -hide_banner -nostats -i read.wav -af volumedetect -f null - 2>&1 | grep max_volume

Say the first read gives max_volume: -2.8 dB. The middle of the target band is -9 dB, so lower the input by about 6 dB and measure again.

A cleanup chain in the right order, with starting values

Every value below is a starting point to adjust by ear, not a standard. Compression lifts quiet material, noise included, so noise reduction comes before it.

Warning: Noise reduction is the step most often overdone. Audacity's manual says higher settings than necessary damage the audio you want to keep, and that sensitivity set too low leaves artefacts it calls "musical noise".

FFmpeg can run steps 1 and 5 plus normalisation; EQ, noise reduction and compression stay in your editor. We ran these on FFmpeg 8.1.1, using a text-to-speech clip rather than a human recording:

# 1. High-pass at 80 Hz, copy the mono voice to both stereo channels,
#    then normalise to -16 LUFS with a -1.5 dBTP ceiling
ffmpeg -i take.wav -af "highpass=f=80,pan=stereo|c0=c0|c1=c0,loudnorm=I=-16:TP=-1.5:LRA=11" -ar 48000 cleaned.wav

# 2. Check integrated loudness (I) and true peak on the file you will upload
ffmpeg -hide_banner -nostats -i cleaned.wav -af ebur128=peak=true -f null - 2>&1 | sed -n '/Summary/,$p'

Per FFmpeg's loudnorm documentation, single-pass use runs in dynamic mode and upsamples to 192 kHz, hence -ar 48000. It lands near the target, not on it: our speech clip read -16.1 LUFS, and a take ending in a loud burst measured a true peak of -0.9 dBFS against the -1.5 ceiling. Check both numbers with command 2 and leave a little margin.

Always measure the file you will upload, after the channel layout is final. Copying one mono channel to both sides at full level reads about 3 LU louder: on our test clip a mono file at -16.6 LUFS read -13.6 LUFS after the pan filter. FFmpeg's documentation calls a mono file's EBU R 128 measurement "perceptually incorrect" for stereo playback, and editors split mono across two channels differently, so measure the final file.

Loudness targets: what the standards say, and what YouTube does not

LUFS (loudness units relative to full scale) is a perceptual loudness measure defined in ITU-R BS.1770; the EBU notes LKFS means the same thing. "Integrated" means averaged over the whole programme, the figure to check on a finished video. True peak (dBTP) accounts for peaks between samples.

Note: The figure usually quoted for YouTube, -14 LUFS, comes from third parties. Ian Shepherd's write-up of the Stats for nerds panel (2017, about music, page updated 2022) describes what he found: loud uploads are turned down, quieter ones are not turned up, and the "content loudness" reading shows how far an upload sits from YouTube's reference level. He stresses it is not about aiming for -14 LUFS.

Our pick for tutorials: -16 LUFS integrated, true peak no higher than -1.5 dBTP. It matches Apple's figure, sits below the number third parties quote for YouTube, and leaves a margin under the -1 dBTP ceiling that EBU R 128 and Apple both use. Keep one number for every video so viewers are not reaching for the volume between tutorials.

To see what YouTube does with your own upload:

1. Upload the video as unlisted.
2. Right-click the player and choose Stats for nerds, the path the 2017 write-up describes. If you cannot find it, YouTube may have moved the panel.
3. Read the content-loudness value. Per the write-up, a negative value means YouTube leaves the video alone, a positive one means it is louder than the reference and gets turned down, and no value means it has not been measured yet, so check again later.

A five-minute test recording that tells you what to fix

Run this before any session that matters, and whenever the mic, desk or room changes. Silence comes first and the clap last, so each can be measured without the other:

[10 seconds of silence, mouth closed]
Peter picked a peck of pickled peppers. Bob brought the big blue bag back.
She sees six sunny seashells on the shore.
And this is the one setting that you must not skip.
[One hand clap]

1. 0:00, pre-flight (30 seconds). Headphones on, notifications silenced, fans off, towel under the mic.
2. 0:30, level (60 seconds). Read the script aloud and adjust gain until the loudest peak lands between -12 and -6 dBFS.
3. 1:30, take (60 seconds). Record the full script with your real recorder, settings and posture, saved as take.wav.
4. 2:30, measure (60 seconds). In a Mac or Linux shell, save the script below as audiocheck.sh and run sh audiocheck.sh take.wav.
5. 3:30, listen (90 seconds). Play it on headphones, then on laptop or phone speakers, closer to what many viewers use.
6. 5:00, decide. Match what you heard to the table, fix one thing, repeat only that check.

#!/bin/sh
# usage: sh audiocheck.sh take.wav [room_start room_len speech_start speech_len]
f=${1:?usage: sh audiocheck.sh take.wav [room_start room_len speech_start speech_len]}
rs=${2:-1}; rl=${3:-8}; ss=${4:-11}; sl=${5:-10}
vd() { ffmpeg -hide_banner -nostats -ss "$1" -t "$2" -i "$f" -af volumedetect -f null - 2>&1; }
room=$(vd "$rs" "$rl" | sed -n 's/.*mean_volume: \(.*\) dB/\1/p')
speech=$(vd "$ss" "$sl")
mean=$(echo "$speech" | sed -n 's/.*mean_volume: \(.*\) dB/\1/p')
peak=$(echo "$speech" | sed -n 's/.*max_volume: \(.*\) dB/\1/p')
echo "speech peak: $peak dBFS (target -12 to -6)"
echo "room tone:   $room dB"
awk -v a="$mean" -v b="$room" 'BEGIN { printf "noise gap:   %.1f dB\n", a - b }'
ffmpeg -hide_banner -nostats -ss "$ss" -t "$sl" -i "$f" -af ebur128=peak=true -f null - 2>&1 |
  sed -n '/Summary/,$p' | grep -E '^ +I:|^ +Peak:' | tr -s ' '

The script measures two windows, not the whole file, because the clap is often louder than your loudest word and a whole-take average is not room tone. -ss is the start and -t the length, in seconds; the defaults assume silence from 1 to 9 seconds and speech from 11 to 21, so check your waveform, pass your own times, and keep the clap out of both. Speech peak maps straight onto the -12 to -6 target. Room tone is the silence's mean level, a root-mean-square (RMS) figure per FFmpeg's volumedetect filter. Noise gap is the speech's mean level minus the room tone, so bigger is better. The last two lines are the speech window's integrated loudness (I) and true peak (Peak).

Our rule of thumb for the gap: 40 dB or more is excellent, 30–40 dB is fine for tutorials, and under about 30 dB means fixing the source or the distance before any software cleanup. ACX's audiobook limits (speech at -23 to -18 dB RMS, noise floor at most -60 dB RMS) imply a gap of at least 37 dB, by our arithmetic.

Match what failed to the cheapest fix. Cost is a class, not a price, because prices move.

A mic also captures conversations and sounds you never meant to publish, so run the screen recording privacy checklist alongside this test.

Where to start this week

1. Run the five-minute test today with the gear you own and note which row fails.
2. Make the cheapest fix for that row, then repeat only that check.
3. Pick one loudness target and use it for every video.

More walkthroughs are in the tutorials hub.

TechCast Studio, our recording workspace, runs readiness checks, audio among them, before you record, and it is in Founding Beta on Apple Silicon. The method above works with any recorder.

Frequently asked questions

Do I need an audio interface or an XLR mic?

A USB mic needs no interface; an XLR mic does. Per Shure, a condenser needs phantom power or a battery and a dynamic mic does not. For a first purchase, fix the room and the distance before upgrading the mic, because a better mic cannot undo an echoey room.

What sample rate, bit depth and channel layout should I record?

Record 48 kHz, 24-bit, one channel of voice. YouTube's recommended upload settings list AAC-LC, Opus or Eclipsa Audio, stereo layouts and 48 kHz, so export the mono voice to both channels, as the first command above does. Apple's podcast requirements also recommend 24-bit for WAV and FLAC.

Why does my tutorial sound quieter than other YouTube videos, and should I fix it?

Possibly, depending on where your upload sits. Per Ian Shepherd's 2017 write-up on music, YouTube turns uploads louder than its reference level down and does not turn quieter ones up. A tutorial at -16 LUFS therefore plays at its own level, a couple of decibels below videos mastered at the commonly quoted -14 LUFS, and pushing a master above the reference gains nothing. A very quiet export, say -23 LUFS, would sound quieter than most videos, and raising it toward -16 helps. YouTube can change this, so check Stats for nerds on your own video; Stable volume, a viewer setting that is on by default, also evens out loud and quiet moments, though not for every video.

Sources and further reading

• Good Sound, Good Research: How Audio Quality Influences Perceptions of the Research and Researcher — Newman and Schwarz, Science Communication 40(2), 2018 (DOI 10.1177/1075547018759345). The two-experiment design and headline result, from the abstract.
• Microphone Techniques for Recording — Shure, undated educational publication. Dynamic versus condenser, spoken-word placement, polar patterns, the 6 dB distance rule and critical distance.
• Recording vocals at home: microphone and room tips — Sonarworks, 2025. Peak range, high-pass and 200–400 Hz advice, the clap test, closet advice and the 30–60 cm critical distance.
• EBU R 128: Loudness normalisation and permitted maximum level of audio signals — European Broadcasting Union, version 5, November 2023. The -23 LUFS target and -1 dBTP true-peak limit.
• Audio requirements — Apple Podcasts, read October 2026. The -16 LKFS figure, tolerance, true-peak limit and format recommendations.
• YouTube Stats For Nerds - EXACT volume normalization values revealed, and how to find them — Ian Shepherd, Production Advice, 2017 (page updated 2022). How content loudness is reported and why quieter uploads are not boosted; written about music, so check your own player.
• Recommended upload encoding settings — YouTube Help, read October 2026. Audio codec, channel and 48 kHz sample-rate recommendations.
• Q. How much headroom should I leave with 24-bit recording? — Hugh Robjohns, Sound On Sound, September 2010. The -18 dBFS reference level and typical peaks around -10 dBFS, for music.