A balanced creator mix is not three tracks set to three magic numbers. It is a hierarchy that changes with the content: speech carries the explanation, music carries continuity and tone, and sound effects clarify selected actions or spaces. The mix works when the intended foreground remains intelligible without making everything else feel absent.
Build that hierarchy in a fixed order. Repair the voice, establish a static balance, remove conflicts, automate changing moments, and only then measure and limit the complete programme. This order prevents a common loop in which every new level change creates another problem.
Define the foreground before touching faders
Write one sentence for each section: “The viewer must hear ___.” In a tutorial, that may be an instruction and the real sound of a mechanism engaging. In a montage, music may be the foreground while effects punctuate two decisive movements. In an interview, speech normally remains primary and the score supplies shape between ideas.
Create separate tracks or buses for:
- dialogue and voiceover;
- music;
- designed effects;
- production sound or ambience;
- final mix processing.
This routing is not administrative decoration. It lets you lower music without damaging speech, compare all effects together, and meter the final signal separately from its components. Label sources and keep unprocessed originals available.
Set a repeatable listening condition
Choose a comfortable, consistent monitoring volume and do not raise it each time the mix feels dull. Louder playback can make a weak balance seem more exciting while tiring your ears. Take short breaks and return at the same volume.
Begin on your best available headphones or speakers, but do not approve there alone. Later you will test a phone speaker, ordinary earbuds and a low-volume pass. Those are translation checks, not separate mixes to optimize independently.
Meters provide repeatable evidence; listening decides whether the hierarchy works. ITU-R BS.1770 defines algorithms for programme loudness and true-peak measurement. Those measurements help compare versions and detect peak risk, but they do not tell you whether a word is masked by a guitar or whether an effect communicates the right action.
Prepare voice as the reference
Solo the speech and edit obvious distractions before adding music. Remove unusable takes, choose clean room tone for gaps, and use short fades to avoid clicks. Address excessive noise or resonance conservatively; heavy processing can make consonants watery or unstable.
Level phrases so the speaker feels consistent. Clip gain or region gain is usually a better first move than extreme compression because it corrects the source before dynamics processing. Use compression only to control the remaining range, not to make every syllable identical. Check plosives, breaths and sibilance in context rather than deleting all human texture.
Listen for meaning. Names, numbers, qualifications, safety instructions and calls to action need special attention. If one word disappears, raising the entire voice track may not solve it; a local gain move, cleaner edit or reduced competing sound is more precise.
Establish a static balance
Bring in music at its busiest representative section, not only under a quiet introduction. Lower it until the speech is easy to follow at your fixed monitoring volume. Then bypass the music and re-enable it. You should hear its character and movement without needing it to prove its presence on every phrase.
Add ambience next. It should maintain continuity and space, not duplicate broadband noise already captured with the voice. Add designed effects last, one at a time, and classify their function:
| Effect role | Example | Mixing priority |
|---|---|---|
| Informational | click confirms a latch closed | audible at the relevant action |
| Orienting | room tone establishes a location | continuous but behind the foreground |
| Transitional | whoosh bridges a deliberate change | brief emphasis without masking the next word |
| Decorative | extra impact on a graphic | first candidate to remove |
If several sounds serve the same role, choose the clearest one. Stacking impacts often produces more peak level and less definition.
Remove masking before adding volume
When voice and music compete, identify where and when the conflict occurs. A continuous level problem needs a different repair from a single lyric or instrument covering one word.
Try these moves in order:
- Arrangement: use a less dense music section, an instrumental version or a deliberate gap.
- Timing: move a fill, effect or musical entrance away from a critical word.
- Level: lower only the competing passage.
- Frequency: make a modest, auditioned reduction in the part of music that masks the voice.
- Dynamics: use controlled ducking or side-chain compression when the conflict follows speech consistently.
Do not apply a copied “voice frequency” preset. Speakers, microphones, instruments and recordings differ. Sweep-and-boost demonstrations can exaggerate problems; make small reductions while listening to the whole mix, then bypass the processing.
Stereo placement can create space, but it is not a substitute for level control. Phones may reproduce a narrow image or a single combined channel. Check the mix in mono so that phase relationships and center-heavy content do not create a surprise.
Automate the story, not every syllable
A single music level rarely fits the whole video. Use volume automation to shape entries, exits, dense sentences, pauses and visual sections. Start a music reduction before the first important word and return gradually enough that the listener does not hear the fader working unless the move is intentionally dramatic.
Automatic ducking can accelerate this pass. Adobe's current Premiere documentation says its ducking feature generates editable gain keyframes based on tagged audio types, sensitivity, reduction amount and fade settings. Treat those keyframes as a draft. Check false triggers from breaths, short pauses and effects; regenerating automation can also overwrite manual changes in the documented workflow.
Keep music present through harmless pauses when rapid pumping would distract. Conversely, lower it more decisively under a quiet or unfamiliar speaker. Automation follows intelligibility and narrative emphasis, not a uniform waveform shape.
Place sound effects around meaning
Line an effect up with the perceptual moment, which may not be the first frame of movement. A closure sound belongs where the object visibly seats; a pass-by may begin before the subject enters frame; a room change may need ambience to lead the cut.
Soloing helps find timing, but approve effects in the full mix. Short transients can look small in a waveform while sounding startling. Use fades where needed, preserve useful production sound, and avoid an effect on every transition. Silence around a key sound can create more definition than added gain.
If an effect is necessary to understand the video, include its meaning in captions. W3C guidance states that captions convey synchronized speech and relevant non-speech audio. Decorative music does not need a constant textual description, but a meaningful alarm, knock, crash or musical stop may need one.
Use accessibility guidance with correct scope
W3C Technique G56 describes a mix in which foreground speech is at least 20 dB above background sound. It is a sufficient technique tied to the Level AAA criterion for certain prerecorded, primarily speech, audio-only material. It is not a universal social-video loudness target.
Still, it provides a valuable stress test: people who are hard of hearing can have greater difficulty separating foreground speech from background. If your creator video depends on spoken information, test a version with substantially lower music or no background at all. Captions complement intelligible audio; they do not excuse an unnecessarily obscured voice.
Measure the complete mix last
Once the balance works, measure integrated programme loudness, short-term behavior and true peak with a standards-based meter. Use the current delivery specification for the actual client or platform; do not inherit a broadcast, podcast or streaming number merely because it appears in a preset.
EBU R 128, for example, recommends -23 LUFS programme loudness for its broadcast context and uses loudness range and maximum true peak as additional descriptors. That is useful evidence that delivery control involves more than one number, not a recommendation to master every creator video to -23 LUFS.
Apply final limiting conservatively and listen again. If the limiter is working hard throughout, return to individual peaks, stacked effects or excessive bus gain. A limiter is the final guardrail, not the tool that creates hierarchy.
Run four translation passes
Export the real delivery format and check it from beginning to end:
- Low volume: Can you follow every critical word without leaning in?
- Phone speaker: Does speech remain clear, and do essential effects survive without harshness?
- Headphones: Are edits, noise changes, stereo distractions or aggressive sibilance exposed?
- Mono and visual-only: Does the audio remain stable when summed, and do captions preserve essential meaning when sound is unavailable?
Write time-coded fixes. “Music too loud” is vague; “lower music under the product warning at 00:18 and lengthen the return fade” is actionable. Recheck the preceding and following seconds after every automation change.
Rights and release record
Confirm permission for music, effects and third-party recordings in the intended commercial context. The U.S. Copyright Office distinguishes a musical composition from a sound recording; permission involving one does not automatically settle the other. Save the exact asset, source, license terms, acquisition date, attribution requirement and project use. Also record whether editing, client work, paid promotion and platform distribution are allowed.
Release only when speech is understandable, meaningful effects are identifiable, music changes support the story, captions cover essential audio, measured delivery meets the current destination specification, and the rights record is complete. The goal is not to make every element loud. It is to make every element's role unmistakable.
Related resources
Sources and scope
- ITU — Recommendation BS.1770: Algorithms to measure audio programme loudness and true-peak audio level — primary technical standard; accessed 2026-09-15.
- EBU — R 128: Loudness normalisation and permitted maximum level — primary broadcast recommendation; accessed 2026-09-15.
- W3C WAI — Technique G56: foreground speech and background sound — primary accessibility guidance; accessed 2026-09-15.
- W3C WAI — Captions/Subtitles — primary accessibility guidance; accessed 2026-09-15.
- Adobe — Automatically duck audio in Premiere — official product documentation; accessed 2026-09-15.
- U.S. Copyright Office — Musical Compositions and Sound Recordings — official authority; accessed 2026-09-15.
The hierarchy, diagnostic order and translation checklist are Angry Media editorial synthesis. EBU R 128 and W3C G56 are presented in their stated contexts, not as universal platform presets. Product interfaces and delivery specifications can change. This article is educational, not legal or hearing-health advice. Source review completed 2026-09-15; scheduled publication 2026-11-04.
