ONE MELODY. ANOTHER DIMENSION.
Find your
second voice.
Sing a little. Choose a harmony. Blend the two.
Your sound stays on this device.
01 Bring a melody
MONO IN · HARMONY OUTOne voice or instrument, up to 60 seconds. Headphones recommended.
Off by default. Applies to the next recording or live microphone session. Stop before changing it. Browser support varies.
The latest take is saved only in this browser for reload recovery. Forget take removes it.
Live microphone · harmonize as you sing
Use headphones to prevent feedback. Choose your Voice, key source and style below before starting. Harmony settings are held during a session; stop to change them. Blend and output levels remain adjustable. No uploaded take or Generate step is needed.
SwiftF0 uses an older pitch estimate (roughly 200–300 ms plus device processing). Audio keeps playing; harmony may lag at note changes. First use downloads about 4 MB. Applies to microphone harmony; fixed octave mode needs no tracker.
Not running. Latency estimates will appear here.
Estimated pitch over time
Measured pitch from your take, with note names and octaves on the vertical axis. The line preserves slides and vibrato; gaps mean no reliable estimate. Tap the graph or move the slider to inspect a moment.
A4 = 440 Hz. This shows tracker estimates from the loaded take, not generated harmony or a live microphone display. Optional take normalization applies before analysis.
Pitch deviation from the nearest note
Estimates from your take, before key filtering, with normalization only if enabled. Reference tuning: A4 = 440 Hz. Cents show how sharp (+) or flat (−) a pitch is. This is not a key estimate.
All notes and octaves share one histogram. 0 ct is the nearest note; left is flat, right is sharp. Bins are 5 cents wide, with the center spanning −2.5 to +2.5 ct. The outer bins stop at −50 and +50 ct. This measures tuning relative to the nearest note, not the intended melody.
Chosen chord timeline
Automatic harmony suggestions, timed from the input. Playback includes additional DSP delay.
Estimated key timeline
Major/minor profile matching is a first approximation. Short or ambiguous melodies may never yield a key. A single octave voice follows the melody while the key is unknown (above low notes, below high notes). Once several notes support a first approximation, two chord voices take over using a tentative key. Recent notes can revise it; stronger evidence establishes it. Key changes still require sustained support. Manual key lock bypasses this wait. The key and scale controls still choose the reference melody.
Octave-down voice selected.
The reference plays the selected key. Loading the test melody replaces your take with generated notes in that key; then generate harmony as usual.
Scale harmony follows one voice in the key you choose. Automatic triads use the selected style, recent notes, chord function, and voice leading; fixed placements are also available. Notes outside the scale and uncertain pitches fade the harmonies out.
Applies to Inspect pitch and Generate voice. SwiftF0 loads about 4 MB of compressed files on first use; audio stays on this device. Live microphone has its own detector setting. Harmonize as it plays uses YIN.
Normalization off. The take uses its recorded level.
Live file processing requires HTTPS or localhost; offline generation works over HTTP.
02 Make the blend
OCTAVE DOWNEach voice can reach 100% (unity gain). Loud blends clip at full scale. Take preview bypasses these sliders; normalization applies only when you enable it.
Play or export to check clipping at the current voice levels.
Ready when you are. Record or upload a melody.