Audio Equalizer — Shape the Frequencies

Almost always the lowest one. Speech recorded in a room picks up rumble below 120 Hz — traffic, air conditioning, the desk knocking the mic — and none of it is audible as sound, only as mush that eats headroom. Pulling that band down by 6 to 12 dB cleans up more than any other single move. Note that this band fades in rather than switching on at 120 Hz: at −12 dB it measures −12 dB at 30 Hz but only −6 dB at 100 Hz, so pull it further than the number suggests. Once the low end is out of the way, a small lift in the 2–8 kHz band is what makes a voice sound present rather than distant.

Below 120 Hz is rumble and handling noise. 120–500 Hz is the body of a voice — too much is boxy, too little is thin. 500 Hz to 2 kHz holds the vowels and most instruments and rarely wants big moves. 2–8 kHz is consonants and presence, the band that makes speech intelligible over music. Above 8 kHz is air and sparkle, and also hiss, fan noise and sharp S sounds.

The tool builds one FFmpeg firequalizer curve from your five values, and the filter smooths every edge over roughly the same ±100 Hz. At 500 Hz, 2 kHz and 8 kHz that is a small slice of a wide band, so those edges behave as steps: measured with a band at −12 dB, the level is still the full −12 dB 100 Hz below the edge and back to flat 100 Hz above it. The 120 Hz edge is the exception, because ±100 Hz down there covers more than two octaves — the same −12 dB setting measures −12 dB at 30 Hz, −11.4 dB at 50 Hz, −8.6 dB at 80 Hz, −6 dB at 100 Hz, −1.5 dB at 150 Hz and nothing at 200 Hz. So the sub band is a gentle low shelf, not a cliff. Where the edges really are steps, setting two neighbouring bands more than about 10 dB apart can ring on percussive material — drum hits, plosives — and spreading the change over two bands avoids it.

Not with this tool. Lifting a band raises peaks, and a track already mastered near 0 dBFS has no headroom to absorb them, so a brick-wall limiter is inserted after the EQ whenever any band is above 0 dB. It holds peaks at −0.26 dBFS — a sliver under full scale, which leaves room for the overshoot an MP3 or AAC decoder can add — and its auto-level is switched off, so the limiter never raises the overall level of the file behind your back. When every move is a cut there is nothing to catch, so no limiter is added and the audio is left alone apart from the EQ itself.

It is stream-copied: the exact same picture data is written into the output file, with no re-encode and no quality loss. Only the audio track is decoded, equalised and re-encoded. Untick the video option and you get an MP3 of the equalised audio instead. MP4, MOV, M4V and MKV keep their container and get AAC audio; WebM gets Vorbis, because the Opus encoder is not usable in this WebAssembly build. Anything else — AVI, FLV, WMV — has to be repackaged as MP4, and the codecs those usually carry cannot be copied into MP4, so untick the video option for them. A file the browser calls video but that holds only sound, such as a WebM from a browser voice recorder, comes back as audio and the tool shows an audio player.

It is the same idea with far fewer controls. A DAW gives you arbitrary band counts, adjustable Q, shelving and parametric curves, and a spectrum analyser to aim with. This is five fixed bands at ±12 dB, which covers the corrective jobs — rumble out, presence in, hiss down — without installing anything or uploading your file. For surgical work like notching a single resonant frequency, use a real parametric EQ.