k2k audio X-ray vision for audio. Separate a sound into its ingredients — transients, harmonics, noise, formants — so you can work on each layer independently.
Every extractor splits your audio into two streams: Extracted (the layer you targeted) and Rejected (everything else). Toggle Invert to swap them.
Transient Extractor
What it does Detects and isolates the attack portion of each hit, pluck, or onset in your audio.
When you’d reach for it You want to soften harsh attacks without dulling the sustain, or grab just the snap of a snare to layer elsewhere.
Quick example
- Feed a drum loop into Transient Extractor.
- Set Duration to 20 ms and Sensitivity around 60%.
- Route the Extracted output (attacks only) through a Smooth node to tame them.
- Blend the softened attacks back with the Rejected output (sustain).
Parameters
| Parameter | What it controls | Range | Sweet spot hint |
|---|---|---|---|
| Duration | Length of the attack window captured around each onset | 10 — 100 ms | 20—40 ms for drums, 50—80 ms for plucked strings |
| Sensitivity | How easily onsets are detected (higher = more hits caught) | 0 — 100% | 40—60% for clean tracks, 70%+ for busy mixes |
| Interval | Minimum gap between detected transients | 10 — 500 ms | 50 ms prevents double-triggers on flams |
| Fade Out | Gradually taper each transient window instead of a hard cut | On / Off | On for natural results |
| Invert | Swap Extracted and Rejected outputs | On / Off | — |
HPSS Extractor
What it does Splits audio into three layers: Harmonics (pitched, sustained tones), Percussives (attacks and broadband hits), and Residual (noise, breath, ambience).
When you’d reach for it You need a clean separation of melody from rhythm, or you want to isolate the room tone hiding underneath a performance.
Quick example
- Load a full mix into HPSS Extractor.
- Select Harmonics to grab vocals and sustained instruments.
- Process the harmonic layer with pitch correction or reverb.
- Recombine with the Rejected output (percussive + residual).
Parameters
| Parameter | What it controls | Range | Sweet spot hint |
|---|---|---|---|
| Component | Which layer to extract: Harmonics, Percussives, or Residual | 3 choices | — |
| H Kernel | Temporal smoothing width for harmonic detection (must be odd) | 11 — 51 | 31 for general use; lower for fast-changing pitched content |
| P Kernel | Spectral smoothing width for percussive detection (must be odd) | 11 — 51 | 31 for general use; lower for narrow-band percussion |
| Power | Mask sharpness — higher values give harder separation | 1.0 — 4.0 | 2.0 for soft Wiener masks, 3.0+ for near-binary splits |
| Invert | Swap Extracted and Rejected outputs | On / Off | — |
Noise Floor Extractor
What it does Isolates continuous, unpitched noise — hiss, room tone, tape noise, breath — from pitched and transient content.
When you’d reach for it You recorded a field recording with steady background hiss and want to pull it out, either to remove it or to use it as a texture.
Quick example
- Feed a vocal recording into Noise Floor Extractor.
- Set Threshold around -55 dB and Smoothing to 50%.
- The Extracted output contains the breath and room hiss.
- Flip Invert on to get a cleaner vocal on Out1 instead.
Parameters
| Parameter | What it controls | Range | Sweet spot hint |
|---|---|---|---|
| Threshold | Magnitude level below which content is considered noise | -80 — -20 dB | -60 dB for light hiss, -45 dB for heavy noise |
| Smoothing | Temporal smoothing of the noise mask (reduces flicker) | 0 — 100% | 40—60% for steady noise; lower for pulsing artifacts |
| Invert | Swap Extracted and Rejected outputs | On / Off | — |
Formant Extractor
What it does Detects and isolates vocal formants (F1 through F4) — the resonant peaks that give vowels their character.
When you’d reach for it You want to reshape a vocal’s vowel quality without touching its pitch, or extract just the “body” frequencies of a voice for layering.
Quick example
- Feed a vocal take into Formant Extractor.
- Select All to capture every formant region, or pick F1 for vowel height only.
- Set Width to 200 Hz for broad capture, or narrow to 80 Hz for surgical work.
- Route the Extracted output through a Gain or EQ node to reshape the vocal character.
Parameters
| Parameter | What it controls | Range | Sweet spot hint |
|---|---|---|---|
| Formant | Which formant(s) to extract: F1, F2, F3, F4, or All | 5 choices | All for general reshaping; F1+F2 individually for vowel design |
| Width | Bandwidth around each formant center frequency | 50 — 500 Hz | 100—200 Hz for vocals; wider for instruments with broad resonances |
| Strength | Intensity of the extraction mask | 0 — 100% | 100% for clean separation; lower for partial blending |
| Order | Cepstral analysis resolution (higher = more detail in envelope) | 8 — 60 | 16 for speech; 30—40 for singing with vibrato |
| Invert | Swap Extracted and Rejected outputs | On / Off | — |
Spectral Envelope Extractor
What it does Separates an audio signal into its timbre (spectral envelope) and its pitch content (excitation), based on source-filter theory.
When you’d reach for it You want to transplant the timbre of one sound onto the pitch of another, or isolate pure pitch content for analysis.
Quick example
- Feed a cello recording into Spectral Envelope Extractor.
- Select Envelope to get the cello’s tonal color without its melodic content.
- Use a Blend node to apply that timbral shape onto a synthesizer sound from another branch.
Parameters
| Parameter | What it controls | Range | Sweet spot hint |
|---|---|---|---|
| Output | Which component to extract: Envelope (timbre) or Excitation (pitch) | 2 choices | — |
| Order | Cepstral resolution — lower values give a smoother envelope | 20 — 80 | 40 for general use; 20—30 for broad timbral shape; 60+ for detailed formant tracking |
| Iterations | Refinement passes for the envelope estimate | 5 — 20 | 10 for most material; 15—20 for sounds with sharp spectral peaks |
| Invert | Swap Extracted and Rejected outputs | On / Off | — |
Spectral Peak Extractor
What it does Finds and isolates the N strongest resonant peaks in the frequency spectrum.
When you’d reach for it You have a resonant instrument and want to grab specific overtones for processing, or you need to notch out ringing frequencies.
Quick example
- Feed a piano recording into Spectral Peak Extractor.
- Set Peaks to 8 and Width Hz to 30.
- The Extracted output contains only the strongest overtones.
- Flip Invert to remove those resonances from the original instead.
Parameters
| Parameter | What it controls | Range | Sweet spot hint |
|---|---|---|---|
| Averaging | How peaks are measured across time: Mean, Median, or Max | 3 choices | Mean for sustained tones; Max to catch brief resonances |
| Peaks | Number of peaks to extract | 1 — 20 | 3—5 for dominant partials; 10+ for a full harmonic portrait |
| Width Hz | Frequency width of the extraction window around each peak | 10 — 200 Hz | 30—50 Hz for precise overtone targeting; 100+ for broad resonance bands |
| Thresh dB | Minimum magnitude for a peak to qualify | -60 — 0 dB | -40 dB ignores low-level noise peaks |
| Invert | Swap Extracted and Rejected outputs | On / Off | — |
Temporal Envelope Extractor
What it does Extracts the slow-moving amplitude shape (dynamics contour) from each frequency band, separating it from fast fluctuations like tremolo or flutter.
When you’d reach for it You want to capture the ADSR-like dynamic shape of a sound to apply it elsewhere, or strip away micro-variations to get a steadier tone.
Quick example
- Feed a bowed violin phrase into Temporal Envelope Extractor.
- Set Smoothing to 80 ms and mode to RMS.
- The Extracted output is the smooth, sustained bow movement.
- The Rejected output contains the vibrato and bow noise detail.
Parameters
| Parameter | What it controls | Range | Sweet spot hint |
|---|---|---|---|
| Type | Envelope calculation method: Mean (smooth), Max (preserves attacks), or RMS (perceptually balanced) | 3 choices | RMS for most material; Max when you need to preserve transient peaks |
| Smoothing | Window size for the moving average | 1 — 500 ms | 30—80 ms for musical dynamics; 200+ ms for very broad contours |
| Invert | Swap Extracted (envelope) and Rejected (detail) outputs | On / Off | — |
Pitch Class Extractor
What it does Isolates every occurrence of a chosen musical note — or an entire chord — across all octaves at once.
When you’d reach for it You want to pull all the C notes out of a piano recording regardless of octave, or extract just the notes belonging to an Am7 chord from a complex mix.
Quick example
- Feed a guitar strumming recording into Pitch Class Extractor.
- Switch to Chord mode, set the root to G, and choose Major.
- The Extracted output contains only the frequencies belonging to a G major chord.
- Process the rejected output separately to reshape the non-chord tones.
Parameters
| Parameter | What it controls | Range | Sweet spot hint |
|---|---|---|---|
| Mode | Single note or Chord selection | Single / Chord | — |
| Note | Which pitch class to extract (C through B) — acts as root in Chord mode | 12 notes | — |
| Chord | Chord quality: Maj, Min, D7, M7, m7, Dim, Aug (visible in Chord mode only) | 7 types | — |
| Tuning | A4 reference frequency | 430 — 450 Hz | 440 Hz standard; adjust for vintage or non-standard tunings |
| Width | Pitch tolerance — how much spectral blur around each target note | 0 — 100% | 10% for precise extraction; 30%+ for looser, more forgiving grabs |
| Invert | Swap Extracted and Rejected outputs | On / Off | — |
Micro Transient Extractor
What it does Detects and isolates ultra-short events under 10 ms — clicks, pops, vinyl crackle, digital glitches.
When you’d reach for it You have a recording with vinyl clicks or digital pops and want to grab them for removal, or you want to harvest those tiny artifacts as a percussion layer.
Quick example
- Feed a vinyl transfer into Micro Transient Extractor.
- Set Duration to 3 ms and Sensitivity to 80%.
- The Extracted output is a stream of isolated clicks.
- Flip Invert to get the cleaned audio on Out1.
Parameters
| Parameter | What it controls | Range | Sweet spot hint |
|---|---|---|---|
| Duration | Length of the capture window around each micro-event | 1 — 10 ms | 2—5 ms for clicks; 8—10 ms for slightly longer pops |
| Sensitivity | Detection threshold (higher = more events caught) | 0 — 100% | 60—80% for typical vinyl; lower for sparse glitches |
| Interval | Minimum gap between detected events | 1 — 50 ms | 10 ms prevents merging adjacent crackle |
| Invert | Swap Extracted and Rejected outputs | On / Off | — |
Silence Extractor
What it does Finds quiet or silent regions in your audio based on an energy threshold.
When you’d reach for it You want to isolate the gaps between phrases for creative filling, or trim dead air from a long recording.
Quick example
- Feed a spoken-word recording into Silence Extractor.
- Set Threshold to -55 dB and Duration to 200 ms.
- The Extracted output contains only the pauses.
- Flip Invert to keep just the speech with silences removed.
Parameters
| Parameter | What it controls | Range | Sweet spot hint |
|---|---|---|---|
| Threshold | Energy level below which a frame counts as silent | -80 — -20 dB | -60 dB for clean recordings; -45 dB for noisy sources |
| Smoothing | Softens the edges of detected silent regions | 0 — 100% | 30% for gentle fades; 0% for hard cuts |
| Duration | Minimum length a quiet region must last to count as silence | 10 — 1000 ms | 100—200 ms skips brief pauses; 500+ ms catches only long gaps |
| Invert | Swap Extracted and Rejected outputs | On / Off | — |