When narration and music sound clean separately but distort or compete when combined, measure and listen to the combined section: adding audio streams can create a higher peak, and the music can mask speech even when nothing clips. Record each track’s peak and the mix peak over the same passage, then adjust the tracks or their timing and listen again. Do not approve the mix just because every solo track stays below full scale.
Start where the problem occurs
Find a short passage where narration, music and any sound effects overlap, especially a loud word, musical accent or impact. Use the same start and end points for every measurement; comparing different sections can make the peak readings misleading. Note whether distortion is audible, whether words are hard to understand, or both. These are different symptoms and may need different changes.
Measure the narration, music and effects separately, then measure the actual combined output. Use the same peak measurement method and units for every track and the mix, over the same timeline passage; readings made with different peak types or scales are not directly comparable. Record the method and units alongside each reading. FFmpeg’s astats can report audio statistics, while amix combines audio inputs. Check the actual filtergraph and mix options used: results depend on the signal path being measured. See the FFmpeg audio filters documentation and its volume filter documentation.
Compare the tracks with the mix
Follow these steps in your editor or a diagnostic workflow using FFmpeg:
- Mark the overlap passage and note its exact duration or timeline range.
- Solo each track in turn. Record the peak reading for narration, music and each effect that overlaps the passage. Also note any crackle or flattening that is already audible in solo playback.
- Play the full mix and record its peak over the same passage. Listen for a peak indicator reaching its limit, audible harshness, and words that disappear under the music. A meter identifies level; it does not decide whether speech is intelligible.
- If the combined peak is the problem, lower the gain of the track or tracks contributing at that moment, or change their automation so the loudest overlap is less crowded. If words are masked but there is no clipping, lower or reshape the music during speech, or adjust its arrangement. Avoid raising the whole mix to fix an isolated balance problem.
- Replay the same passage with narration, music and effects together, then replay a nearby section where one of those elements enters or leaves. Record the new readings and whether the important words remain clear.
FFmpeg’s volume filter can apply a gain change to an audio stream, and amix can combine streams. These provide a way to inspect or process a mix, but they do not replace a listening decision in the editing context. A filtergraph’s routing and mix options affect what reaches the output, so verify that the measured stream is the actual intended mix rather than a different input or intermediate bus.
Diagnose the type of failure
The following hypothetical readings show how to separate the likely causes. Values are illustrative dBFS peak readings from the same passage, not measurements from a real project or recommended targets.
| Track or check (all readings use the same peak method, units and passage) | Hypothetical observation | What it suggests | Next move |
|---|---|---|---|
| Narration solo | Peak at -4 dBFS; no audible roughness | No obvious solo overload in this passage | Keep it as a comparison point |
| Music solo | Peak at -3 dBFS; clean on its own | Music may still add to the mix peak or mask speech | Check its level during spoken words |
| Combined mix | Peak reaches 0 dBFS; roughness on a coincident accent | Suspect overload in the combined path | Reduce or automate contributing tracks, then remeasure |
| Combined mix | Peak at -2 dBFS; a name is hard to understand | Peak headroom remains in this reading, but masking is possible | Rebalance music or arrangement and listen again |
| Narration solo | Same roughness is audible before mixing | The problem may already be in the narration source or its processing | Inspect the source and upstream processing; mix gain cannot restore missing clean audio |
A clean solo reading does not rule out clipping after summing. Conversely, a low combined peak does not prove dialogue is clear: competing sounds can obscure words without pushing the meter to its limit. Treat the meter and the audition as separate evidence.
Recheck the full overlap
After an adjustment, check more than the single loudest instant. Play the whole dialogue phrase with music and effects at their actual timeline levels. Listen once at a comfortable working level for intelligibility, then repeat at a quieter playback level to check whether essential words remain distinguishable. If a particular consonant, name or short line still disappears, adjust the balance or arrangement around that phrase rather than applying a global gain increase.
If the combined meter still reaches its limit, inspect routing and any bus or output processing before assuming an individual source is responsible. If distortion is audible in a solo track, lowering its fader may reduce its contribution but will not necessarily remove distortion already present in that source or an upstream stage. Avoid adding a limiter as a substitute for identifying which track, bus or stage is overloaded. The provided FFmpeg documentation describes filters and their options; it does not establish a universal safe peak target or guarantee that a gain change will make every mix clear.
Questions creators ask
Why can the mix peak higher than every individual track?
Overlapping waveforms add at the mixing stage. Their combined value at some instants can exceed the peak of either input, depending on their levels and timing. Check the peak after the actual mix, not only on isolated inputs.
If the mix does not clip, why is the narration still hard to hear?
A peak meter reports level, not whether speech is easy to distinguish from music and effects. Listen to the words in context and change the balance or arrangement if they are masked.
Should I reduce every track by the same amount?
Not necessarily. First identify which sounds overlap at the problem moment and whether the failure is overload, masking, or source distortion. A targeted adjustment can address the cause while preserving other scene elements.
Sources
Sources and next step
- FFmpeg Filters Documentation
- 8.35 amix — FFmpeg official documentation
- 8.120 volume — FFmpeg official documentation
This guide interprets public documentation and uses explicitly hypothetical acceptance examples, not application tests. Controls and availability can change; verify the cited manual against your installed version.
For work created in Imagild, carry the input requirements and acceptance checklist into the film workspace. The external software controls discussed above belong to their respective products.