The studio work, done by asking for it.
Most tools give you a rack of plugins and wish you luck. dhun gives you a producer. You bring a track, a stem or a voice memo, say what is wrong with it, and it does the work — then tells you what it did and why.
Free to start · No card · We take no rights in your music
It is a conversation, not a menu.
You do not need to know that the fix is a 3 dB dip at 2.8 kHz with a medium Q. You need the vocal to be audible.
here’s the rough mix — the vocal is buried and the room is really obvious
Analysed it. The vocal is masked between 2–4 kHz by the guitars, and there is 50 Hz hum plus broadband room noise. De-noising first, then I will carve the guitars and lift the vocal rather than just turning it up.
can you make it sit like the reference i uploaded
Matched to it — tonal balance and loudness. Yours was 3.2 LU quieter and had more energy below 100 Hz. Have a listen; I can back off the low end if it feels thin now.
the third line is flat
Line three, "and I never came back" — D3 sung about 40 cents flat across two words. Corrected just those, formants preserved. The rest of the take is untouched.
Illustrative exchanges, in the product’s own voice. The analysis they describe — masking pairs, band energy, cents-flat detection — is what the engine actually reports.
Eighteen tools, and it picks them.
Every one of these runs on our servers, not in your browser — which is why the phone app can do the same work as the desktop.
Separate the stems
Vocals, drums, bass and the rest, pulled apart from a finished mix. Runs Demucs server-side, so it works the same on a phone as on a laptop.
split_stems
Tune a vocal so it still sounds human
Pitch is detected with CREPE and corrected with rubberband, which preserves formants — the reason a corrected take keeps its voice instead of turning into a chipmunk. It can also list what you actually sang, note by note, and fix only the ones you name.
tune_vocal · list_vocal_notes · fix_vocal_notes
Remove the room
DeepFilterNet takes out hiss, hum, air-conditioning and street noise without the underwater artefacts a gate leaves behind.
denoise
Balance the mix, with reasons
It analyses the whole mix — band-by-band energy, crest factor, sibilance, phase correlation, and which tracks are masking each other — and then tells you what it changed and why.
analyze_mix · mix_balance
Master it, or match a record
A Pedalboard mastering chain for a straight master, Matchering when you want it to sit like a specific reference track, and pyloudnorm to hit a platform LUFS target exactly.
master_track · normalize_loudness
Write something from nothing
A full track from a prompt, or a single instrument part — a bassline, a pad, an arpeggio — dropped onto the timeline in the key and tempo of what is already there.
generate_song · generate_instrument
Transcribe what was sung
Lyrics out of a vocal take, timed, ready for captions or a lyric video.
transcribe
Render and export
The full timeline with every insert chain applied, bounced server-side at full quality.
render_mixdown
92 plugins. You never have to open one.
The producer knows what each of these is for and reaches for the right one — an opto compressor on a vocal, a bus compressor on a drum group. They are all there if you want to drive them yourself, and unlimited on every plan including free, because they cost CPU rather than vendor money.
Dynamics
15opto, FET and bus compressors, multiband, transient, gate, clipper
Time
14plate, hall, spring and shimmer reverbs, tape and ping-pong delays
EQ
12vintage, dynamic, mid-side, Baxandall, 31-band graphic, notch
Saturation
12tube, transformer, diode, fuzz, wavefolder, bitcrusher
Utility
11pan, phase, polarity, mono-maker, M/S matrix, metering
Modulation
11chorus, flanger, phaser, rotary, tremolo, ring mod
Mastering
7master EQ, bus comp, stereo enhance, limiter, dither, reference match
Creative
7granular, glitch, lo-fi, sidechain
Pitch
3correction, doubler, octaver
A real DAW, for when you want one.
Multi-track timeline, insert chains, automation, offline render. It is a genuine editor, not a toy.
It is also optional and secondary, on purpose. The editor came first and got too complicated for the people it was for. Everything it can do, the producer can do by being asked — so the editor is there when you want to move a fader yourself, and out of the way when you do not.
Web only. It uses Web Audio for live playback, which is a browser technology — the phone apps do the same work server-side instead.
Multi-track timeline
Arrange, trim, fade, automate.
Insert chains
Any of the 92, in any order, per track.
Full-quality render
Offline bounce, exactly what you hear.
We take no rights in your music.
Nothing you upload or make here becomes ours. We do not take a cut, we do not claim a share of publishing, and we do not need a licence to your masters to run a compressor on them.
We are also straight with you about the part that is genuinely unsettled: for audio you upload and we process, the result is yours. For audio generated from a prompt, copyright is not ours to promise — the US Copyright Office does not register work lacking human authorship, and the generation providers grant a commercial licence rather than ownership. Anyone telling you otherwise is making a promise they cannot keep.
Read how we handle AI-generated audioBring a track and ask it something.
Free to start, and the free plan is counted rather than timed — one song, three stem splits, one master, twenty replies. No clock, no card.
Start free