Audio leveler free, in your browser
Bring the quiet voice up and the loud one down, so a conversation sits at one steady level. It works phrase by phrase on your device, leaves the pauses quiet, and saves a WAV or MP3.
Processing runs on your device. Up to 200 MB and 60 minutes. Nothing is uploaded.
What to expect
- Cost and privacy
- Free, with no account. Processing runs on your device; your audio is never uploaded.
- Use it for
- Even out speech that changes volume inside one mixed file: a host and a quieter guest on one track, a speaker who leans in and out, or a call where one side is quieter.
- Inputs and output
- WAV, MP3, M4A and other files your browser can decode. Saves a 16-bit WAV or a 192 kbps MP3 at -16 LUFS with peaks under -1 dBTP.
- Limits
- Up to 200 MB and 60 minutes, subject to available browser memory.
- Keep in mind
- It hears quieter stretches of speech, not voices. When two people talk at once both move together, and noise under the quiet speaker comes up with their words.
How to level speech volume, in four steps
- STEP 01
Choose a recording
Drop a WAV, MP3 or M4A file, or try the sample conversation with a quiet guest. It decodes and levels on your device; nothing is uploaded.
- STEP 02
Read the finding
The screen says how far the quiet speech sits behind the loud speech, then shows the spread of short-term loudness before and after leveling.
- STEP 03
Set the Amount
Amount starts at 100%. Lower it if the host's emphasis sounds flattened; raise it if the quiet voice still drops away. Each change levels the original again.
- STEP 04
Compare and download
Play Before and After from the same spot, then save a 16-bit WAV or a 192 kbps MP3 at -16 LUFS with peaks under -1 dBTP.
What a leveler does
A leveler is an automatic hand on the volume fader. It listens to speech a few seconds at a time, finds the phrases that sit below or above the middle of the conversation, and turns them up or down toward that middle. The ride is capped at 12 dB either way and moves slowly, over about a quarter of a second, so the rise and fall of words inside a sentence stays where it was. Amount scales the whole ride: at 50% each phrase moves half as far.
Pauses are not speech, so they are not turned up. When a gap lasts longer than about half a second, the leveler holds the volume at or below where it was on either side. Room tone and breaths between turns stay quiet instead of rising with the quiet guest.
After the ride, the whole file is set to -16 LUFS, a common loudness for spoken podcasts, and a true-peak limiter keeps peaks under -1 dBTP. The finding first says in words how far apart the quiet and loud speech sit, then shows the spread of 3 second loudness from the quietest tenth of the speech to the loudest tenth, before and after.
Leveler, track gain or normalizing?
- Use the leveler when the volume changes over time inside one mixed file: a host and guest recorded to one track, a speaker who leans in and out, or a call where one side is quieter.
- Use track gain in the audio editor when each person already has their own track. One steady change per track is cleaner than a ride, and it leaves every breath where it was.
- Use the quiet audio fixer when the whole recording is too quiet but evenly so. It lifts the speech as a whole and keeps a gap between speakers mostly in place.
- Normalizing sets a file to a loudness target with one gain change. Check the result in the LUFS meter. It does not close a gap between voices; the leveler closes the gap and then sets the loudness.
Why normalizing does not balance host and guest plays the same conversation normalized, track-gained and leveled, with each speaker measured on their own.
Where it struggles
- Crosstalk. When two people talk at once, it hears one phrase and moves both voices together. It cannot pull one voice up out of the other.
- Laughter, coughs and shouts. To a leveler these are loud speech, so a long laugh can be turned down a few dB along with the words around it.
- Music beds and sound effects. Loud music counts as speech, so a bed under the talk gets ridden with the voices. Level the voice before adding music.
- Noise. Turning a quiet speaker up turns up the hiss under their words too. Treat the noise first, then level.
- Long files take time. It works on the whole file, up to 60 minutes and 200 MB, and an hour of audio can take a minute or more on a slower computer.
Questions
Is the audio leveler free?
Yes. Leveling and the WAV or MP3 download are free, with no account. The audio is processed in your browser and never uploaded.
How is leveling different from normalizing?
Normalizing moves the whole file by one amount, so a guest 9 dB behind the host is still 9 dB behind afterward. A leveler moves each phrase on its own, so the quiet voice comes up and the loud one comes down. This tool then sets the whole result to -16 LUFS, so you get both.
Can I balance a podcast host and guest recorded on one track?
Yes, that is what it is for. It does not recognise voices; it hears a quieter stretch of speech and raises it. If the host and guest are on separate tracks, a steady gain change per track in the audio editor is cleaner.
Will it turn up background noise in the pauses?
Not in pauses longer than about half a second. The ride is held at or below its level on either side of a gap, so the room does not swell between sentences. Noise under the quiet speaker's words does come up with them.
Does it work on music?
It is tuned for speech. On a song it would ride the quiet verses up toward the chorus and flatten the arrangement. Use a mastering tool or plain normalization for music.