Agree the destination before the session
A delivery specification should distinguish original recordings, edited masters and implementation assets. These are not necessarily the same files. Capture may preserve multiple microphone perspectives, while the game needs a selected mono event. An editor may need long handles and unprocessed alternatives even when the final engine asset is tightly trimmed. Clarify each stage before recording begins.
Ask the receiving team for a small approved example. A written specification can leave room for different interpretations of naming, silence, channel order or loudness. Delivering a handful of test files early exposes those differences cheaply. Import them into the intended workflow and confirm they behave correctly before processing a whole session. A successful export is not proof of a successful handoff.
Choose formats for an actual reason
Uncompressed PCM WAV is a common exchange format, and 48 kHz, 24-bit recording is a common production choice. Neither is a universal requirement. Higher sample rates may help particular sound-design transformations, while adding storage and processing demands. Agree the capture format with the people responsible for editing and implementation rather than treating larger numbers as an automatic quality improvement.
Broadcast Wave can carry useful metadata within a WAV-compatible structure; EBU Tech 3285 describes that format. Metadata support varies between applications, so verify which fields survive your chosen route. Preserve essential information in a readable manifest as well. A human should be able to understand the delivery without depending on one specific workstation or an undocumented database.
Keep identifiers stable and names useful
Use a naming convention that carries the minimum information required to find and distinguish an asset. Stable line or event identifiers are usually more valuable than long prose descriptions. Include versions deliberately, and avoid silently replacing an approved file with a different performance under the same name. If an identifier changes, document the relationship between old and new references.
Sound-effects libraries may benefit from the Universal Category System, often called UCS, where its taxonomy fits the receiving team's needs. Dialogue may instead follow project-specific line IDs and locale codes. Do not impose one naming scheme on every asset simply for visual consistency. Whatever scheme you choose, test its characters, length and sorting behaviour in the actual delivery environment.
Edit without erasing future choices
Keep source recordings separate from derived edits. Remove obvious technical distractions only to the extent agreed, and retain unprocessed alternatives when restoration or creative processing is substantial. Aggressive denoising can produce moving artefacts that become obvious after further compression or spatialisation. An apparently clean solo audition may hide problems that surface in the finished mix.
Leave intentional room for attacks and decays. Short fades can prevent clicks, but a fade that softens a transient may change the action. For loops, test the boundary repeatedly at the intended playback behaviour. For dialogue, preserve natural consonant beginnings and enough context for an editor to make a deliberate transition. Tightness is useful only when it does not damage the performance.
“The next team should not have to reverse-engineer what the session already knew.”
— Roverbravilo editorial principle
Check both sound and structure
Listen for clipping, missing beginnings, unwanted edits, inconsistent tone and truncated tails. Then check filenames, channel counts, durations and the correspondence between files and the manifest. Automated checks can identify technical exceptions, but they do not replace listening. Equally, a careful listening pass may not notice that two otherwise valid files share an identifier and overwrite one another during import.
Do not apply a broadcast loudness target to every isolated effect without a project reason. ITU-R BS.1770 defines loudness and true-peak measurement algorithms, while EBU R 128 provides a broadcast loudness recommendation. Neither automatically specifies the correct gain for every game asset. Engines such as Audiokinetic Wwise and FMOD Studio introduce additional routing and runtime behaviour that must be considered in context.
- Source recording
- The preserved capture before editorial selections and delivery processing.
- Manifest
- A readable inventory linking filenames to identifiers, descriptions, versions and relevant recording notes.
- Checksum
- A value used to verify that a file's contents have not changed during copying or transfer.
Make recovery part of completion
Maintain independent copies in separate failure domains and verify them rather than assuming a synchronised folder is a complete backup. Agree access, retention and deletion rules, especially for confidential scripts and identifiable voices. Include a concise readme explaining folder structure, known issues and outstanding decisions. Completion means another authorised person can locate, understand and recover the material without needing the original engineer beside them.