Build the sound architecture first
Microdrama sound has at least five layers that are independent and compete for attention: dialogue, narration, music, ambience and Foley. Pushing them all into one "generate the complete audio" task loses versioning, editability, rights provenance and mix control.
Every sound asset gets a stable ID, a text or music version, its generation or recording parameters, a speaker or theme, timecode, rights provenance and approval state. The edit timeline references assets; the final audio is never treated as a black box you cannot take apart.
Musical consistency comes from a grammar of motifs
A season should neither loop one song nor swap in a random new track each episode. Define short motifs for core characters, relationships or secrets first, then state how they may vary:
- tonality: bright, ambiguous, threatening, resolved;
- tempo: lurking, driving, chasing, stalled;
- instrumentation: identity and relationship ownership;
- density: solo, sparse pulse, full ensemble;
- endings: open, suspended, deceptive cadence, resolution;
- stems: rhythm, low end, harmony, melody and texture controlled separately.
The music bible records where a theme first appears, which characters may share a variant, which scenes forbid it, its maximum intensity and its licensing boundary. That way different episodes sound like one world while still reacting to the state of the story.
The cue sheet is music's execution contract
Each cue records at minimum its entry, its exit, its story function, the emotional start and end, the theme and variant, the intensity, where it conflicts with dialogue, its loopable region, its stems and its provenance. Not "put some tense background music here."
Music often should change state before the picture cuts: the low end enters first, and only then does the audience begin to sense danger. A reversal may call for an instant emptying rather than another layer of drums. Precise cues are finished once picture durations stabilize, but the thematic design has to enter much earlier, at the story stage.
Produce long narration by meaning and breath
Do not chop two minutes of narration mechanically every 200 words. Build a prosody map first: for each unit of thought, its stress, tempo, pauses, emotion, breathing and relationship to the picture. Put the divisions at the boundaries of thought, with overlapping context either side, so a model or performer knows how the previous unit ended and where the next is going.
Lock voice, sample rate, loudness target, microphone character, sense of space, tempo range and pronunciation lexicon across the whole thing. Keep a dry master for each segment before de-noising, de-essing, breath work and equalization. Prefer joins at natural pauses, either side of a plosive, or where ambience can mask them — and record every seam.
The approved narration master becomes the master clock for picture and subtitles. Changing one word later can change breathing, duration, cut points, music cues and subtitles, so it goes through a change request rather than a quiet audio swap.
How dialogue, narration and music coexist
- Make the most semantically important sound intelligible first.
- Have music avoid key consonants and sentence-final information, not just drop in overall level.
- Use stems to reduce instruments sharing frequencies with the voice, instead of flattening the whole cue.
- When narration enters, keep enough spatial cues in the ambience so it does not become a vacuum voice-over.
- Re-listen on a phone speaker, on ordinary headphones and in a noisy environment, checking comprehension at low volume.
Rights and temp music
Store music provenance, licence, order, download date, territory, platform, term and whether advertising use is permitted. Temp music is marked conspicuously and replaced before picture lock. AI-generated music also records the model, the terms, the prompt, the output and any similarity review.
The standard for finished sound is not that it sounds full. It is that the audience understands the information without effort, feels the emotion, and believes that adjacent shots share one space and one time.