apulSoft splitS v1.1.2 [WiN]

• Product: splitS
• Developer: apulSoft
• Version: 1.1.2
• Format: VST3, AAX
• Requirements: Windows 10 or later
• Source: apulsoft.ch/splits

splitS is a sibilance isolation processor that uses spectral brightness, level, and pitch analysis to divide a vocal signal into two full-range components — sibilant and non-sibilant — which sum back to the original signal with 100% integrity when no gain change is applied. The S fader and NOS (non-sibilance) fader control the balance between the two components independently. Detection is shaped by three parameters: Quietness (minimum sibilance level threshold), Loudness (maximum non-sibilance level before misclassification), and Tone (spectral brightness sensitivity). S-Range, added in v1.1, caps maximum S-event volume or excludes soft events from the S channel. Multi-out support routes NOS and S as discrete signals in Pro Tools, Reaper, Live, and Bitwig. Typical latency runs approximately 30 ms. It answers the query: how do I reduce vocal sibilance without triggering compression or altering the frequency content of the rest of the signal.

Key Takeaway

Vocal sessions where a de-esser has already been applied but sibilance character still requires manual level correction — or where de-essing artifacts have introduced lisp or compression pumping — are the primary context where splitS changes the result. The fader-per-split-component architecture replaces gain automation on individual sibilant clips with a continuous, detection-driven volume ride. The ~30 ms latency is parameter-dependent and not zero; sessions requiring strictly zero-latency processing on the vocal bus are outside its operating condition. Engineers whose sibilance problems stem from tonal harshness at a specific frequency rather than from relative level between sibilant and non-sibilant segments will get more precise results from a dynamic EQ or multiband tool targeting that frequency range.

Spectral Brightness Split vs. Frequency-Domain De-Essing

A conventional de-esser applies gain reduction to a frequency band — typically a shelf or bell centered in the 5–10 kHz range — when the signal in that band crosses a threshold. The gain reduction applies to the full-range signal through a crossover or to the target band in isolation, which introduces tonal changes to the non-sibilant portion whenever the compressor opens. splitS operates differently: the detection stage classifies each moment of the signal as sibilant or non-sibilant using spectral brightness, level, and pitch simultaneously, then routes the full-range signal — not a filtered version — into the corresponding component. The S component contains the whole signal during classified sibilant moments; the NOS component contains the whole signal during non-sibilant moments. Pulling the S fader down reduces the level of those moments without touching the spectral content of either component, which is why no EQ or crossover filtering artifact appears on the non-sibilant material. The tradeoff is that detection accuracy determines everything — classification errors at the sibilant/non-sibilant boundary produce level transitions at the wrong moments, which require Quietness, Loudness, and Tone adjustment to correct rather than a simple threshold move.

Detection Parameters: Quietness, Loudness, Tone

Three controls shape what the detection engine classifies as sibilance. Quietness sets the minimum level a moment must reach to be classified as an S event — raising it excludes soft sibilances that sit below the threshold, reducing false positives on breath noise and room sibilance. Loudness sets the maximum level that non-sibilant material can reach before the detector potentially misclassifies a loud consonant or transient as sibilance — adjusting it downward tightens the boundary but increases the risk of catching loud non-S consonants. Tone shifts the spectral brightness reference point — a vocal recorded with a bright condenser in a reflective room presents a different brightness baseline than a closer, darker source, and Tone compensates for that before the detector applies its brightness criterion. The three parameters interact: a Quietness setting that works on one vocalist may misclassify breathy sibilances on another, and a Tone setting calibrated for one microphone chain can misread the same voice through a different chain. Calibrating detection on a representative section of the specific vocal before committing to reduction settings produces more consistent results across a session than setting the fader first and adjusting detection to match.

S-Range: Level Ceiling for S Events

S-Range, introduced in v1.1, adds a maximum level cap for classified S events and a minimum level gate that excludes soft S events from the S channel entirely. The maximum level cap prevents individual loud sibilant hits from remaining at full volume after the S fader reduces the overall S channel — without it, a very loud sibilant at -6 dBFS and a moderate one at -18 dBFS both receive the same proportional gain reduction from the S fader, which may leave the loudest hit still audible above the intended ceiling. The minimum level gate removes soft S events — breath sibilances, room decay with sibilant character — from the S channel classification, leaving them in the NOS channel where the non-sibilant fader governs their level instead. The combination of these two range boundaries narrows the classification to the specific S events that need attention while leaving the softest and loudest edges in their natural signal path. S-Range is not a transient limiter applied after the fact; it reshapes which events enter the S classification before the fader acts on them.

Multi-Out Routing and the AudioSuite Path

Multi-out support is available in Pro Tools, Reaper, Ableton Live, and Bitwig Studio, routing the NOS and S components as discrete output signals to separate tracks. In this configuration, the NOS track and S track exist as independent audio streams in the session timeline, allowing per-clip gain automation, different processing chains on each component, and region-based edits on the sibilant track without touching the non-sibilant material. On S events that need heavier treatment than the S fader alone provides — a single phrase where the sibilance is significantly louder than the rest of the vocal — the multi-out path lets an engineer print the S component and edit its gain at the clip level rather than adjusting the global fader. In Pro Tools, a SoundFlow script package captures the insert detection settings and transfers them to the AudioSuite version, allowing offline rendering of selected regions at different S fader values than the insert is set to — useful when a section requires more reduction than the rest of the session without a separate instance. The multi-out path is host-dependent: DAWs that don’t support multi-output plugin routing cannot access discrete NOS and S channels and receive only the mixed stereo output.

What splitS Doesn’t Resolve

The detection stage classifies by level and spectral brightness over time — it has no understanding of phonetic content. A loud non-sibilant consonant with high spectral brightness, a sharp transient on an acoustic guitar string, or a cymbal hi-hat pattern with sibilant character in a bus context can register as S events and receive the S channel’s gain reduction. These classification errors require Loudness and Tone adjustment to push the boundaries around the target material, and on complex or polyphonic sources the detection may not be stable enough to classify selectively. The plugin was designed for single vocal tracks and performs most predictably there; on bus use, acoustic guitar string noise, and cymbal content the detection boundaries require more calibration and the results are less consistent across different material. The ~30 ms latency is typical and parameter-dependent, meaning it requires delay compensation in any session where timing precision with other tracks matters. Producers who need a sibilance solution that operates at true zero latency throughout the signal chain — live monitoring paths, strict sample-accurate tracking sessions — need to account for that delay before placing splitS in the signal chain.

FAQs