How low should background music be under a voice?

Every editor eventually meets the same complaint from a viewer: I can't hear what they're saying over the music. And every editor has also heard the opposite: the video feels dead — where's the music?

The fight between background music and speech is as old as film sound, but there are working numbers for it. This article collects the levels practitioners actually use, explains why your mix sounds different on a phone than in your editing headphones, and covers the rescue path when the voice was recorded too quiet under the music and you can't re-record.

The working numbers

Sound engineering doesn't have a law for background music, but it has strong conventions:

• Speech peaks around −6 dBFS on your meter is a common target for talking content — loud enough to sit comfortably, with headroom to avoid clipping.
• Background music around −18 to −25 dBFS under speech is the standard ducked range for dialogue-heavy content. Some practitioners quote a simpler mnemonic: voice at −6, music at −30 for "music barely there," landing in the −15 to −20 range for energetic vlogs.
• A difference of roughly 15–20 dB between voice and music is the zone where speech stays effortlessly intelligible for most viewers.
• Outside of narration-heavy work (like a documentary mix at −30), music-only sections of the video can come back up to full level between sentences — ducking is dynamic, not a static fader position.

One practitioner rule worth stealing from cooking-video creators who tested this on real audiences: if you think the music is finally quiet enough, it's probably still one or two steps loud. Cut it again and re-check on the worst speaker you own.

Why your mix lies to you in the studio

Three reasons a mix that sounds balanced on your setup fails on a phone:

1. Small speakers can't reproduce the low end. Music's bass and pads mostly vanish from a phone speaker — but speech doesn't. So music that seemed politely quiet in your headphones loses exactly the part that was masking it, and what remains of the midrange sits on top of the voice.
2. Compression you didn't notice. Streaming platforms normalize loudness; quiet passages get pushed up, loud ones pulled down. The voice-to-music ratio you carefully set can be re-balanced by the pipeline.
3. The masking curve. Human hearing masks quieter sounds that share frequency bands with louder ones. A voice at −6 and music at −20 feels fine in a quiet room — and fails on a bus. Always check the mix at low volume, on the worst device.

The rescue path when the voice was recorded too quiet

Now the hard case: the footage already exists, the music is baked into the track, and the voice under it is too quiet to rescue with an EQ or a volume fader — because they're one mixed audio track. You can't change the ratio of two things that are one signal.

This is where separation changes the game. AI separation splits the single track into two independent layers — the voice and the music — and once they're separate, you can do what was impossible before:

1. Drop the video into a separation tool. The video file stays on your device; only the extracted audio track (a few MB) is processed, and deleted within 24 hours.
2. Download the voice-only video. The picture is untouched — the video stream is copied, never re-encoded — and the voice is now free of the music that was burying it.
3. Rebuild the mix properly: voice at full presence, replacement music ducked 15–20 dB below, using the level numbers above. If you kept the music stem, you can even reuse the original song at the correct level where rights allow.

Step 2 is the unlock. Editors used to solve "voice too quiet" by re-recording the narrator or sweating over EQ notches; separation gives you a clean vocal layer from footage you already have. With BGM Remover the whole rescue is one drop and one download — a one-time 5-minute free trial covers testing it on your hardest clip first.

A practical checklist before publishing

• Voice meter peaks ≈ −6 dBFS; music ducked 15–20 dB under speech.
• Music returns to full level only where nobody is talking.
• Check the mix at low volume on a phone speaker — the worst case is the real case.
• Hard-to-hear words survive compression and masking; borderline ones don't. If you're reading lips to verify a line, the music is still too loud.
• If the source audio is unrecoverable (voice truly buried), separate first, fix the ratio after — don't ship a mix you already doubt.

Frequently asked questions

Is −30 dB for music too quiet?

For documentary and dialogue-critical content, no — that's the convention for "music you feel, not hear." For vlogs and high-energy edits, −15 to −20 dB under voice is common. The number matters less than the phone-speaker test.

Can EQ alone fix buried voice?

Only marginally. Cutting the music's midrange frequencies helps a little, but when voice and music share one track, EQ reshapes the mix — it can't un-mix it. Separation is the tool for an already-baked track.

Does separation damage the voice quality?

Good separation leaves clean speech mostly untouched; artifacts show up more in the music stem than the voice. Dense mixes are harder than clean ones. Test a short excerpt — a real free trial exists for exactly this.

What about music that swells over the voice on purpose?

Artistic ducking (music swelling at the emotional beat, then back down) is a stylistic choice — but the voice must stay intelligible during the swell. The 15–20 dB rule applies at every moment voice is speaking, not just on average.

Fix the buried voice

If your footage has great speech trapped under loud music, you don't need a re-record — you need the layers split. Separate the voice from your video, rebuild the mix with the numbers above, and publish a version viewers don't have to strain to hear.