Why your game's sound gets tiring: game audio mixing with volume tiers, ducking and three more tricks

Every sound in your game is fine on its own, but after ten minutes of play the whole thing is tiring, and the warning that mattered got lost under the music. That’s a mix problem, not a sound problem. Our ears are most sensitive between about 2 and 5 kHz (Wikipedia), so piled-up bright sounds wear people out fast. Overwatch’s sound team aimed for players to “play by sound” with 12 players and their weapons all making noise at once (GDC Vault).

Each demo below plays the same 4-second moment on both sides, without the trick on the left and with it on the right. Sound is off until you press Play, and only one demo plays at a time. Auto A/B lets you hear one loop without, then one loop with; tap a side’s chart to listen to just that side. Under each side you see a level chart, a live meter with a CLIP light, and a “Measured” line from rendering the exact loop offline. Try it yourself buttons fire the sounds by hand. Every sound is generated in code.

1. Volume tiers

Rank every sound and give each rank its own volume. Key cues (danger, hits on the player, pickups) play loudest, actions sit about 6 dB lower, and music and ambience sit 12 to 18 dB lower. Route each group into its own bus, so you set one number per group instead of fixing sounds one by one. Overwatch reportedly uses an “importance system” that raises the sounds of your biggest threats “while dimming everything else” (PC Gamer). In our demo, the alert and pickup sit +9.8 dB over the music without tiers and +24.8 dB with them. The Music and Effects sliders only change the right side and are saved between visits. Give players separate sliders like these; the Game Accessibility Guidelines ask for separate controls for effects, speech and music (Game Accessibility Guidelines).

2. Ducking

When a line of dialogue or a key cue plays, lower the music for a moment, then bring it back. Ducking means one signal turns down because another one is playing (Wikipedia). The music never stops, so it doesn’t feel like a glitch. In our demo the music drops 9 dB in about 50 ms and comes back over about 400 ms. The voice sits +8.4 dB over the music without ducking and +17.3 dB with it. Press Say a line and listen for the words on each side.

3. Voice limit

When many copies of the same sound start together (coins, shots, footsteps), they add up into one loud smear. Cap each sound at 3 to 4 copies at once, cut off the oldest copy to make room, and skip repeats that come within about 50 ms. Our demo also plays each new copy 2 dB quieter for each copy already playing. FMOD Studio has these as built-in settings: Max Instances, Stealing and Cooldown (FMOD manual). In our demo, a burst of coin pickups plays up to 12 copies at once on the left and never more than 4 on the right. Try Coin (tap fast!) or Coin burst ×18.

4. High-cut

Take a little off the top. Put a gentle high-shelf cut on the music (about −4 to −5 dB above 6 kHz), and soften the highs of bright sounds that repeat a lot, like shots and coins. In the browser this is a BiquadFilterNode set to highshelf or lowpass (MDN). Our demo uses −5 dB above 6 kHz plus a lowpass at 10 kHz on the music. The share of energy above 6 kHz falls from 34% to 12%. It sounds a little duller on a short listen; the gain is that it stays pleasant after ten minutes. Hold Hold to autofire to hear it.

5. Limiter

End the whole chain with a limiter, so a pile-up never goes past 0 dB and crackles. The browser’s DynamicsCompressorNode “can help prevent clipping and distortion when multiple sounds are combined” (MDN). Ours uses threshold −6 dB, ratio 20, attack 3 ms and release 250 ms. In our demo a boss goes down with explosions, an alert, a pickup and coins all at once. Without the limiter the peak reaches +4.6 dB with 932 clipped samples; with it, −3.7 dB and none. A single explosion on its own loses only about 1 dB. One thing to watch: browsers add hidden make-up gain to this node (about +3.4 dB with these settings), so our demo turns the level back down right after it.

Mix last, and measure

These tricks work together, and each can go too far. Duck too deep and the music pumps; cut too much and the game sounds muffled; limit too hard and loud moments lose their punch. The skill keeps every trick behind its own switch and checks the mix with numbers:

• Render one busy moment with the tricks on and off. An OfflineAudioContext does this in the browser.
• Clipped samples on the master should be zero.
• Key cue over music while the cue plays: our demo went from +9.8 to +24.8 dB.
• Most copies of one sound at once: 3 to 4.
• Energy above 6 kHz: lower with the tricks on.
• Then listen. Numbers catch clipping and pile-ups, but not taste. Play for ten minutes at a normal volume and see if you want to turn it down.

If you still need the sounds themselves, start with adding sound effects with an AI agent.

Add it to your game

1. Copy the skill at the end of this page into your project.
2. Ask your agent: “Make my game’s sound clearer and less tiring.”
3. Play it, then tune in plain words: “The music still covers the warning sound.”

Font in the demos: Atkinson Hyperlegible by the Braille Institute of America (SIL Open Font License 1.1). All sounds are generated in code.