Auto mix: put your vocals on the beat
Drop in a beat, a vocal recording or both. It sets the levels, fixes the tone, compresses, tames sibilance, adds reverb, auto-tunes if you want, and masters a finished track. Every knob is then set for you on a studio-style mixing console, and you keep adjusting each fader, EQ and compressor. Every setting is measured from 144 professional mixes, not guessed. Your audio never leaves your device.
- Your audio stays on your device
- Mixing console, auto-tune, EQ, compressor
- Download WAV at −14 LUFS
1. Drop your files
Beat (instrumental)
Backing track, karaoke or instrumental
Lead vocal
Your vocal recording
Backing vocal
optional
Reference track
optional: a song you like, the tone is matched to it
MP3, WAV, FLAC, M4A, OGG · up to 10 minutes
2. Pick a style
Styles are listening suggestions, not measurements. «Natural» is exactly what the analysis measured.
Drop at least one file: a beat, a vocal, or both.
What it is for
- Put your cover vocals on a karaoke beat and get a finished song.
- Polish a demo recording when you have no mix engineer.
- Process an a cappella take or a voice-over on its own, or master a beat on its own.
- Remix the beat and vocal you got from the vocal remover or AI cover.
How to use
- Drop a beat, a vocal recording or both. Backing vocals and a reference track go in the other two boxes.
- Pick a style (Pop, Ballad, T-Pain…) or leave it on Natural: exactly what the analysis measured.
- Click Auto mix, listen Before / After, drag the sliders until it sounds right, then download the mix. Need every knob? Open the mixing console.
Record in a room with little echo, about a hand span from the mic, and never so loud that the meter hits red. Tone and level can be fixed; room echo and a distorted recording cannot.
Frequently asked questions
- What does the auto mix do?
- In order: cuts hum below 80 Hz, gates noise between phrases (only when it measures a noise floor), auto-tune (if on), vocal EQ, de-essing, compression, reverb, balances the vocal against the beat, gently dips the beat around 2.5 kHz while the vocal plays so the words stay clear, matches the tone to the reference track (if any), then masters to −14 LUFS with a true peak no higher than −1 dBTP. Any step can be switched off.
- What if I only have a vocal, or only a beat?
- Vocal only: the vocal is processed (filter, EQ, de-esser, compressor, reverb, auto-tune if on) and mastered, which suits an a cappella take or a voice-over. Beat only: only loudness and peaks are mastered, the tone is NOT changed, because every genre sounds different and there is no single standard to pull a beat towards. To adjust the tone too, add a reference track.
- What can the auto-tune do?
- It tracks pitch every 10 ms and pulls it to the nearest note, or to the notes of the key you choose or the key it detects. «Gentle» only moves the centre of each note and keeps vibrato; «Hard» forces every moment onto the note, for the robotic sound. Measured on a real vocal detuned by 30 cents: back to 12 cents (gentle) and 6 cents (hard). The price is a slightly rougher voice, so it is off by default.
- What is the reference track for?
- It measures the tone (the spectrum per octave) of the reference and of your mix, then moves your mix towards it, at most ±4 dB per band. Loudness stays at −14 LUFS; it does not copy the loudness of the reference. Pick a song in the same genre.
- What does “measured from real mixes” mean?
- We took 144 songs from the MUSDB18 dataset, where vocals and accompaniment come as separate tracks, and measured three things: how loud the vocal sits against the music, the vocal spectrum per octave, and how much reverb it carries. The EQ only corrects bands where your vocal falls OUTSIDE the range that holds half of those mixes. If your vocal is already fine, it is left alone.
- Why is the vocal louder than the music by default?
- In those 144 songs, mostly Western pop and rock, the vocal sits −2.2 LU below the music (median). A real Vietnamese song we separated and measured had the vocal 2.3 LU above the music. We have few such samples yet, so the default keeps the vocal clear; drag the slider left if you want it to blend in.
- Is my audio uploaded to a server?
- No. Every step runs in your browser on your own device. No copy is sent anywhere, so closing the tab loses it: download your mix first.
- Why −14 LUFS?
- That is the level YouTube and Spotify normalize to. Louder tracks get turned down, so mastering louder only costs dynamic range without sounding louder.
- What if the vocal is out of time with the beat?
- Use the “Offset” slot on the Vocal channel strip, in 10 ms steps. To cut and move sections, click “Open the timeline” after mixing: the processed tracks are loaded so you can drag and trim by hand.
What this tool does
Drop in a beat, a vocal recording, or both, plus a harmony vocal if you have one. The tool measures every file, then filters, shapes the tone, compresses, de-esses, adds reverb, balances voice against music and masters everything into a finished WAV. Every setting it chose appears on a studio style mixing console, so you can keep adjusting or leave it as it is.
All processing runs inside your browser. Nothing is uploaded to a server, so an unreleased recording stays on your computer from start to finish.
The chain, in order
- Low cut on the vocal, with the cutoff set from the lowest pitch of that voice (70 to 160 Hz).
- Noise reduction between phrases, only when real background noise is measured. A clean vocal skips this stage.
- Auto-tune, if you switch it on.
- Octave band EQ: only bands that fall outside the middle range of professional mixes are corrected, and only up to the edge of that range, so every voice is not painted the same colour.
- De-essing and two stage compression, with a little warmth.
- Reverb is tuned step by step until the vocal's stereo width matches the median of real mixes; echo follows the beat's tempo.
- Voice to beat balance, then a gentle dip around 2.5 kHz in the beat while the voice sings, so the words are not buried.
- Master to −14 LUFS with a true peak no higher than −1 dBTP.
Where the targets come from
The targets were measured on a public stem dataset of 144 songs, mixes made by people with every track available separately. In that set the vocal sits 2.2 LU below the music (median; half the songs fall between −4.3 and −1.0). We did not take targets from the site's own AI songs, because that would measure how a machine mixes, not how people mix.
Vietnamese pop tends to put the voice on top: the site's AI songs measured +1.9 and one real Vietnamese song +2.3. So the default puts the voice 2 LU above the music, the slider moves ±4, and a mark at −2.2 shows the dataset median. To be plain about it: the Vietnamese sample is a single song so far.
Four modes and the console
| You provide | What the tool does |
|---|---|
| Beat and vocal | Processes the vocal, balances it with the beat, masters |
| Vocal only | Processes the vocal, adds reverb, masters; harmony sits 6 LU below the lead |
| Beat only | Masters loudness and peak only. The beat is deliberately not auto-EQ'd, since every genre is different |
| Plus a reference track | Matches tone to the song you choose, at most ±4 dB per band; loudness is not copied, it stays at −14 LUFS |
There are seven presets. The natural preset equals exactly the settings the tool chose, while Lo-fi masters quieter, measured at −16 LUFS. On the console, faders, pan, mute and the master target change instantly while you listen; changing a plugin setting rebuilds only that channel. You can compare the mixed version with a raw sum at the same loudness, so a louder version cannot fool your ears into sounding better.
Auto-tune: Light, Medium, T-Pain, and the cost
The tool detects the song's key. Light moves the centre of each note and keeps vibrato; T-Pain snaps pitch instantly for the familiar robotic sound. On a real voice singing about 30 cents off, an independent measuring tool read 12.3 cents after light correction and 6.5 cents after strong correction. The T-Pain setting reached 87%, close to the ideal ceiling of 89% measured on the same test.
The cost is a slightly rougher voice: jitter rises by about 0.2 percentage points and harmonic clarity drops by 0.5 to 0.8 dB. That is why auto-tune is off by default and worth switching on only when the singing is off pitch. On a voice that is already in tune, light mode changes almost nothing.
How long it takes
A 4 minute song with a harmony vocal takes about 10 to 12 seconds to mix on our test machines; the final export adds about 2 seconds. Because everything runs on your device, a slower computer takes longer, but there is never a server queue.
Limitations
- It cannot rescue a distorted or clipped recording; it balances, colours and cleans within reason.
- EQ only fixes bands that are clearly off compared with real mixes. For a specific sound, use a reference track or adjust by hand on the console.
- The voice to music level is based on a dataset made mostly of non Vietnamese music; the Vietnamese sample is one song so far.
- A vocal recorded out of time with the beat is not aligned automatically; there is an offset knob to shift it by hand.
- Files can be up to 10 minutes long.
Frequently asked questions
Is my music uploaded to a server?
No. Every stage runs in the browser and the WAV is built on your device. Close the tab and nothing remains.
I only have a finished mixed song. Can I split it and remix it?
Yes. Use Stem splitter to separate the beat and the vocal, then bring both here.
Is the output ready for streaming platforms?
A loudness of −14 LUFS with a true peak no higher than −1 dBTP is a common level for streaming. The tool's loudness meter matched an independent meter on 10 of 10 test files, within 0.05.
Does the console really change the export, or is it just for show?
It changes the export. In our test, muting the vocal channel and exporting raised the correlation between the export and the beat from 0.723 to 0.994, meaning the file really was the beat without the voice. For more on mixing, see the blog.
Related research
Want to build the mix by hand from scratch?
The music mixer lets you drag, trim, and set the volume and pan of every track.
Open the music mixer