unity-audio

Unity audio

Fundamentals

• One AudioListener per scene (on the camera). AudioSource per emitter; spatialBlend 0 = 2D (UI, music), 1 = 3D (world SFX, set min/max distance + logarithmic rolloff).
• Import settings by use: short SFX → Decompress On Load + PCM/ADPCM; music/ambience → Streaming + Vorbis; many mid-length clips → Compressed In Memory. Force To Mono for 3D-positioned SFX (halves size, panning comes from spatialization anyway).
• AudioMixer from day one: Master → Music / SFX / UI groups; route every AudioSource to a group. Volume settings UI drives exposed mixer params — convert linear sliders to dB: mixer.SetFloat("MusicVol", Mathf.Log10(Mathf.Max(v, 0.0001f)) * 20f). Ducking (music dips under dialogue/SFX): mixer snapshots or a duck-volume effect on the Music group.

Playing patterns

• One-shots: source.PlayOneShot(clip) — fine for UI and sparse SFX. Rapid-fire SFX (gunfire, footsteps, impacts): a small pool of AudioSources (see gamedev-patterns pooling), never one AudioSource per spawned object.
• Kill repetition: randomize pitch ±5–10% (source.pitch = Random.Range(0.95f, 1.05f)) and rotate 2–3 clip variants.
• Music: two AudioSources crossfading, or mixer snapshot transitions; for intro+loop tracks, schedule with PlayScheduled(AudioSettings.dspTime + …) — Invoke/coroutine timing drifts.

Getting audio into the project — four pipelines

0. Code-synthesized (free, no keys, instant) — for retro/arcade SFX and placeholders, write the sound: a Python script using only stdlib wave/math (oscillators + pitch sweeps + envelopes + filtered noise) synthesizes coin/laser/jump/explosion/click WAVs directly into Assets/Audio/SFX/. A working starter generator ships with the plugin: python "${CLAUDE_PLUGIN_ROOT}/scripts/gen-sfx.py" <output-dir> — copy and tune its per-sound functions for the project's aesthetic. Recipes: coin = two-note rising square arpeggio with fast decay; laser = exponential sweep 1800→300 Hz; explosion = white noise through a widening moving-average (darkening tail) with long decay; click = 6 ms sine blip + noise burst. Normalize to ~0.85 peak, 44.1 kHz 16-bit mono. For runtime/variated SFX inside Unity, usfxr (C# sfxr port — sounds are parameter sets, mutate for variation). Local AI models (Stable Audio Open, AudioCraft) are the heavier free tier — GPU + setup required, and verify the model's output-license terms before commercial use.
1. ElevenLabs MCP (bundled as elevenlabs; needs ELEVENLABS_API_KEY set in the user's environment — never ask for the key in chat): text_to_sound_effects (prompt + duration), compose_music, text_to_speech for voice. Always pass an output path inside the project (Assets/Audio/...) — its default output directory is the Desktop. Licensing: free tier is non-commercial + attribution; any paid plan includes commercial use — remind the user before a commercial build ships generated audio.
2. Stock CC0: Kenney audio packs (10 packs, CC0, great UI/impact/jingle coverage), OpenGameArt/Freesound via the asset-scout agent (license varies per asset — scout reports it).

Pick 0 for instant free retro SFX, 3 for stock placeholder/prototype speed, 1–2 for tailored or realistic content. Generated audio lands under Assets/Audio/<SFX|Music|Voice>/; check import settings after (they default wrong for the use case more often than not).

Verify — honestly

You cannot hear. Verify what's verifiable: clip imported with expected length/channels/sample rate (manage_asset), AudioSource wired to the right mixer group with sane spatial settings, no console errors in play mode when the sound triggers. Then ask the user to listen for subjective quality — say explicitly that acoustic quality was not machine-verified.