What this is
seance is a workbench for taking music apart. You feed it one audio file, it draws two pictures of the sound, measures the tempo and the key, and lets you slow the whole thing down, cut pieces out, and save them. A second tool, the acid machine, builds the classic squelchy synth line from scratch so you can compare your imitation against the real record.
Everything happens inside this page on your computer. Nothing is uploaded anywhere — the file never leaves your machine.
First three moves
1. Drop an audio file anywhere on the page (or press Load audio). MP3, FLAC, WAV and OGG all work.
2. Wait a moment — the thin colored line under the top bar is the analysis filling in. When it finishes, the tempo and key panels light up.
3. Press Space to listen. Click anywhere on a picture to jump there. Drag across a picture to mark a region, press L to loop it.
If something breaks
The Log panel at the bottom writes down everything the tool does, with timings. Press Copy log and paste the text to whoever is helping you debug — errors show up in red, and the header of the copy describes your browser and audio setup.
- Space
- play / pause
- Enter
- return to start (or to the loop start)
- ← →
- nudge one beat · hold Shift for four
- I O
- set the selection in / out point at the playhead
- L
- loop the selection
- S
- snap on / off — selections stick to the beat grid
- + − 0
- zoom in / out / fit the whole track
- Z
- zoom to the selection
- Ctrl+E
- bounce the selection to a 24-bit WAV
- ↑ ↓
- change a focused note in the acid sequencer
- ?
- open this guide · Esc closes it
What the waveform shows
Sound is air wiggling. A microphone writes that wiggle down as a line, and the waveform is that line seen from very far away: time runs left to right, and the taller the shape, the louder that moment is. Silence is a flat line in the middle; a kick drum is a sudden spike; a whole song is a mountain range.
Modern loud masters are squashed so hard that the mountain range becomes a solid sausage. That is not a display error — that is how the record was made. If the file is even louder than the format officially allows (it happens), the display scales itself so nothing gets cut off.
The ruler and the grid
The strip above the waveform is a clock — minutes and seconds. After analysis, thin vertical lines appear over the waveform itself: that is the detected beat grid. Every line is one beat; every brighter line is a bar start, the count of one. If the bright lines land on the kick drums, the tempo reading is trustworthy. If they drift against the hits, be suspicious and check the half/double alternatives in the tempo panel.
Touching it
Click anywhere to jump the playhead there. Press and drag to mark a region — the tinted band with edges. The marked region is what looping and bouncing operate on. The × in the status bar below the spectrogram clears it. With snap on (the magnet in the transport, key S), region edges click onto the detected beat grid, so a dragged loop is automatically a clean number of beats — turn it off for free-hand surgical cuts.
Zooming
Hold Ctrl and scroll on either picture to zoom in at the cursor, exactly like a DAW arrangement view; plain scrolling pans left and right. Z zooms to the marked region, 0 fits the whole track, +/− step around the playhead. While playing, the view follows the playhead. The ruler and the beat grid rescale with you — zoom into one bar and place selection edges on individual hits.
What the spectrogram shows
The waveform shows when things happen; the spectrogram shows what pitch they are. Time still runs left to right, but now height means pitch: deep bass lives at the bottom, cymbals and air at the top. Brightness means energy — the hotter a spot glows, the more of that pitch is present at that moment.
The pitch axis is spaced like a piano, not like a ruler: every doubling of frequency (one octave) takes the same vertical distance. The numbers on the left rail are Hertz — vibrations per second. 55 Hz is a low bass A; 440 Hz is the A an orchestra tunes to; 8000 Hz and up is mostly shimmer.
Learning to read shapes
Vertical stripes that span many pitches at once are hits — drums, clicks, anything sudden. Horizontal lines are held tones. One musical note never draws just one line: it draws a ladder — the note itself at the bottom (the fundamental) plus evenly spaced copies above it (harmonics). A voice is a wavy ladder whose rungs get brighter and darker in bands — those bands are the vowels. A synth with a moving filter is a ladder whose upper rungs sweep open and closed.
One forensic tell: if the picture ends in a hard flat ceiling around 16 kHz, the file was lossy-compressed somewhere in its life (MP3, stream rip). A real lossless master keeps faint energy right up to the top.
The inspector
Hover the picture and the status bar reads out: the time, the frequency in Hz, the nearest musical note with its offset in cents (100 cents = 1 semitone, so ±10 c means "basically in tune"), and the level in dB relative to the loudest spot. This is how you identify the pitch of a vocal or a bassline: hover its lowest ladder rung and read the note.
Tempo
BPM means beats per minute — how fast you would nod along. The tool finds it by marking every moment the sound suddenly gets louder (the hits), then asking: what repeating spacing explains these hits best? It assumes four beats per bar, like almost all electronic music, and settles on a number between 90 and 180.
Tempo has a built-in ambiguity: the same groove can honestly be counted fast or slow. 170 with a lazy half-time feel and 85 with busy hats are the same music. That is why the panel always shows half and double next to the main number — for this genre, the half value is usually the one your head nods to. Confidence tells you how strongly the hits agree with the grid; below ~40 %, treat the number as a guess.
Do not trust tempo numbers from websites. Measure. Databases are frequently off by wild ratios because their analyzers count the fastest subdivision.
Key
The key is the home base of a piece: the note where melodies feel resolved, plus a mood — major (brighter) or minor (darker). The tool builds a fingerprint of how much each of the twelve note names is used across the whole track, then compares it against reference fingerprints for all 24 keys and reports the best match with its correlation r (closer to 1 = better).
Every major key has a twin minor key that uses the exact same notes — F major and D minor, for example. Only the emphasis differs, so the two can score almost identically; the panel therefore always shows the relative key next to the winner. If they are close, trust your ear for the mood and take the note set as the fact. And remember: once you screw a track down, everything sounds lower — five semitones of slowdown moves D minor territory toward B♭ minor.
The screw
The screw slider is a tape machine: slowing the tape lowers pitch and tempo together, locked. That linked drop — not a fancy pitch-shifter — is the classic chopped-and-screwed sound this whole genre is built on. The readout shows speed in percent and the pitch drop in semitones: there are 12 semitones in an octave, and −5 st works out to 74.9 % speed. The presets are the standard moves; 45→33⅓ is the original trick of playing a 45 rpm single at 33⅓ (74.07 %).
The effective BPM line does the arithmetic for you: source tempo × speed = what you will actually hear.
Cutting pieces out
Drag across the waveform or spectrogram to mark the piece you want — a vocal phrase, a chant, one bar of drums. Press L and Space to loop-audition it. Then Render WAV: the marked region is rendered through the current screw setting, at the file's own sample rate, and a colored file chip appears — click it to save. No region marked means the whole track exports.
Copy report puts a plain-text summary — file, tempo, key, varispeed, region — on your clipboard for your notes. Bounces are 24-bit WAV at the file's own sample rate: the stem format every DAW expects, ready to drop straight onto an Ableton or Pro Tools track.
Splitting a region
Split selection pulls a marked region apart into two stems using a beautifully simple trick: on the spectrogram, things that persist across time (voices, pads, bass notes, the chant) draw horizontal lines, and things that spike across frequency (kicks, claps, clicks) draw vertical stripes. A median filter run along each direction separates the two, and each half is rebuilt into audio. The harmonic stem holds the tonal material; the percussive stem holds the hits.
The forensic move: mark the vocal or chant, split, then press the small wave button next to the harmonic chip to load that stem back into the editor — now the spectrogram shows the voice alone, and the hover inspector reads its actual notes without drums in the way. Stems are mono, capped at 30 seconds per split, and save as 24-bit WAV.
The other half of the tool
Seance takes music apart; golem puts music together — the same workbench read backwards. It is a step sequencer: three tracks (a 303-style acid lead, a sub bass, and a drum machine) sharing one clock, one root note and one scale. Pick the track with the tabs above the grid; the knobs underneath always belong to the track you are looking at.
Root, scale and length
Notes are stored as scale degrees, not fixed pitches — degree 0 is the home note, 1 is the next note up in the chosen scale, and so on. That is why changing the root from A to D, or the scale from natural minor to phrygian, instantly transposes every track without breaking the pattern. Set the pattern length to 8, 16 or 32 steps; swing nudges every second step late for a shuffled feel.
What an acid line is
One cheap synth (the Roland TB-303) playing one note pattern in fast sixteenth notes, while a resonant filter opens and closes over it — that squelchy, talking, rubber-band sound is "acid", and dark trance builds whole tracks on it. This machine is a browser 303 aimed at that sound.
The sequencer
For melodic tracks, the row of cells is one bar chopped into sixteenths; the numbers mark the four beats. Click a note cell to step through an A-minor scale, drag up and down to sweep it, or focus it and use ↑ ↓. A · is a rest. Below each note: the round LED is accent — that step hits harder and the filter kicks wider open. The diamond is slide — the note glides into the next step without a new pluck, the signature 303 slur. Accents and slides, more than the notes, are what make a line groove.
The knobs, in plain words
- cutoff
- how open the tone gate sits — low is dark and muffled, high is buzzy
- reso
- a whistle right at the gate's edge; push it and the line starts to scream — that is the acid
- env amt
- how far each note kicks the gate open before it falls shut again
- decay
- how fast it falls shut — short is plucky, long is sweepy
- accent
- the extra punch accented steps get
- drive
- fuzz and warmth after the filter
- delay mix / feedback
- the dotted-eighth echo behind the line — mix is how loud, feedback is how many repeats
Drums
Select the Drums track and the grid becomes three on/off lanes — kick, clap, hat. Click cells to build a beat. Witch house convention: put the clap where a snare would go, and let the kick sit heavy and slow.
Bounce → seance
Bounce → seance writes four bars of all three tracks mixed together to audio and drops it on the table like any other file. Two reasons to do it: it proves the tempo and key detectors on a signal whose truth you know, and it gives you a spectrogram of your line to hold next to the record's — matching the filter sweeps by eye is how you close in on the original patch.
Taking the pattern into a DAW
MIDI writes every track into one Standard MIDI File: the lead on channel 1, the bass on channel 2, and the drums on channel 10 with General MIDI note numbers, so they land on a drum rack automatically. Accents become velocity 127, slides become overlapping notes — the convention every 303 plug-in (ABL, Phoscyon, D16) and Ableton's own instruments understand. Drag the .mid chip onto a track and the pattern plays your synths at the set BPM. One thing to know about Ableton's own files: an .als set or .alc clip contains no audio, only references to sample files on disk — so to dissect something from a project, load the referenced WAV/AIFF itself.