The goal of commercial interview audio is not to make every track loud. It is to make every sentence effortless to understand: clear and stable dialogue, believable ambience, music that supports rather than masks, and a final loudness that travels well across publishing platforms.
Diagnose the problem before stacking effects
“Bad sound” can mean changing mic distance, constant noise, room reflections, sibilance, frequency masking from music, or mismatched recording devices. Identify the cause first so that one process does not undo another.
| Issue | Typical symptom | First response |
|---|---|---|
| Dialogue too quiet or uneven | Changing distance to the microphone | Clip gain, keyframes and light compression |
| Constant noise | Air conditioning, electrical or fan noise | Light denoise; preserve a natural floor |
| Muddy voice | Too much low end or a covered microphone | High-pass filter and small low-mid cuts |
| Harsh sibilance | Overly strong “s”, “sh” or “ch” sounds | De-essing or a small high-frequency cut |
| Music masks speech | Listeners need to work to hear dialogue | Ducking followed by manual keyframes |
Organise tracks and sync first
Do not start with effects. Keep the interviewee, interviewer, location ambience, music and sound effects on separate tracks so repair, ducking and final control stay predictable.
- A1 / A2: interviewee and interviewer dialogue;
- A3: location ambience or room tone;
- A4: music;
- A5: supporting effects;
- Bus or mix track: final loudness and peak control.
If you recorded separate audio, sync it by timecode or waveform. Once the edit is structurally stable, begin detailed audio work so that you are not repeating it after every picture change.
Put dialogue into a sensible working range
Use timeline keyframes, Audio Gain or the Audio Clip Mixer to correct obvious level jumps. As a rough working range, interview dialogue can first be normalised to peaks around -6 dB to -3 dB. This is not the final loudness target; it simply gives the mix room to work.
Dialogue chain: repair before polish
A useful order for interview dialogue is denoise → dereverb → high-pass filter → EQ → de-ess → compress → limit. In Premiere Pro, start with Essential Sound for a quick base, then add individual effects only where the recording needs them.
Light denoise and dereverb
In Essential Sound → Dialogue → Repair, begin Reduce Noise at 2 or 3; take the same low-and-slow approach to Reduce Reverb. A small, stable noise floor usually sounds more natural than aggressive cleanup. Over-processing can create watery speech, metallic artefacts, truncated tails and pumping ambience.
High-pass filtering and EQ for intelligibility
| Range | Possible issue | Useful starting point |
|---|---|---|
| 70–110 Hz | Vibration, wind, footsteps or stand noise | High-pass; often start around 70–90 Hz for lower voices and 80–110 Hz for higher voices |
| 120–500 Hz | Mud, boom or boxiness | Small corrective cuts as needed |
| 2–4 kHz | Weak clarity and intelligibility | A small 1–2 dB boost if needed |
| 5–8 kHz | Harshness or sibilance | De-ess or make a small cut |
| 10–12 kHz | Lack of air | Only a very small lift when necessary |
These numbers are starting points, not presets. Microphones, speakers and rooms vary. EQ is mostly problem solving, and a careful cut is often more natural than another boost.
De-essing, compression and limiting
Sibilance often lives around 4–8 kHz; reduce only the harsh consonants. For compression, a useful starting range is Threshold -18 dB to -12 dB, Ratio 2:1 to 4:1, Attack 5–20 ms, Release 60–150 ms, with roughly 3–6 dB of gain reduction. Finish with a limiter around -1 dB peak. If it is working constantly, revisit gain staging and compression earlier in the chain.
Finish dialogue before adding music
The order matters: make dialogue clear and stable, add music, then control overall loudness. During speech, music often sits roughly 12–20 dB below dialogue. It may rise in non-dialogue sections, but should never jump up on the end of an important sentence.
Premiere Pro's Essential Sound can create a first ducking pass: tag voice as Dialogue, tag the bed as Music, enable Ducking on the music track, choose Dialogue as the target, then generate keyframes. Start around -12 dB to -18 dB of ducking with a 200–500 ms fade. Automatic ducking is a first pass; commercial delivery still benefits from manual adjustment around sentence gaps and transitions.
If the music still masks voice even at a modest level, the issue may be midrange competition. A gentle 1–4 kHz cut of 1–3 dB on the music can make room for speech without thinning the whole track.
Let ambience keep edits believable
Do not remove every trace of the room. A controlled bed of ambience fills edits, joins different camera angles and prevents sudden shifts in background noise. Record several seconds of room tone where possible, layer it gently under cuts, and use fades to maintain one coherent space.
Match multiple speakers without making them identical
Interviewers and guests may use different microphones and positions. The goal is not identical tone; it is a natural transition: similar loudness, manageable brightness differences, coherent sense of space and stable panning. Match loudness first, make separate EQ corrections, then keep compression intensity broadly comparable.
Final loudness and delivery checks
For commercial web video, an integrated target around -16 to -14 LUFS is a practical reference, with True Peak no higher than -1 dBTP. Do not flatten all dynamics just to chase loudness. Continuous dialogue intelligibility is more useful than a brief loudness advantage.
A reusable fast template
- Remove obvious clips, impacts and empty mistakes;
- Use keyframes to even dialogue clips;
- Apply light denoise and only necessary dereverb;
- High-pass, EQ, de-ess, then lightly compress;
- Add music and create a ducking pass;
- Manually check music keyframes, edit points and returns between sentences;
- Check overall loudness, peaks, opening and ending;
- Listen on headphones, computer speakers and a phone speaker before export.
Five frequent mistakes
- Only raising volume without controlling dynamics. Loud phrases clip while quiet ones remain unclear.
- Maxing out denoise. It often produces watery voices, metallic artefacts and digital texture.
- Boosting every frequency range. The result becomes harsh and crowded; correct problem ranges first.
- Relying on ducking alone. It creates a useful first pass but cannot replace final human judgement.
- Judging only on headphones. Also check ordinary computer or phone speakers for real-world dialogue clarity.
Takeaway
The priorities are clear: intelligibility before tone, obvious problems before fine polish, and natural sound before audible processing. When dialogue, music, ambience and loudness each have a clear role, an interview mix becomes more stable and translates better between listening devices.