Game audio workflow
Start with an event inventory and native motion/gameplay references. Separate unit voice, movement/gear, weapon air, material impact, victim reaction, UI, strategic alerts, music and ambience. A good recording can still be wrong for the event, camera distance or crowd mix.
Prepare and select material
For each event, record its meaning, motion anchor, source or generation route, desired length and review state. Reuse the user's chosen provider, voice and references. Inspect current tool capabilities instead of copying a stale model identifier. Keep generation IDs to recover pending jobs without duplicating them.
Use stable take IDs and direct local playback. Keep source, audition edit and runtime file distinct. Record cut points, fades, gain/pitch/time changes and source hashes. A compressed preview is not a lossless production master. Review actual samples for clipping, silence, format, duration and loop seams; technical checks do not establish the desired performance or tone.
Keep candidate, liked-for-comparison, accepted, integrated and mix-reviewed states separate. Preserve rejected history without presenting it as selected. Do not normalize or replace an accepted original silently. One accepted source may map to multiple events without duplicate files.
Follow the real event
Split an attack into optional effort, swing air, confirmed material contact and recipient reaction. A cancelled attack produces no contact sound. A death voice is separate from the body's actual landing. Footsteps follow planted-foot events or the project's verified distance/phase system; an arbitrary timer or a full audition loop may drift from the animation.
Play one appropriate speaker for a group command, with per-speaker repetition limits. Repeated path updates may affect movement without repeating the voice. Choose concurrency, variation and distance limits from the actual mix. Coalesce related alerts and fire a match result once from the authoritative result event, not whenever the HUD refreshes. Do not add a parry cue without a real parry event.
Integrate and listen
Use the project's mixer groups and persistent settings. Verify pause, mute, scene changes, interruption, visibility and source cleanup. Audition at gameplay distance and with a representative crowd, then check normal device output. Preserve the user's mix; do not replay unrelated narration during a focused test.
Deliver an event-to-file manifest, provenance/licences, original and derived files, selected states, event anchors, native listening capture and open mix decisions. Name which checks were technical and which were listened to.
Example: a robot's walk, swing, metal impact and shutdown. Successful delivery keeps the impact conditional on a hit, aligns steps, limits crowd repetition and preserves the approved takes. Dependencies: an audio editor or FFmpeg for edits, chosen generation/stock tools when requested, and engine access for integration.
