Sound — Adaptive Audio Plugins for the Voice
Media
2026

Sound — Adaptive Audio Plugins for the Voice

The Clarity Series: adaptive audio plugins that read the signal in real time and tune themselves to it, so one intuitive control does what used to take an afternoon of settings.

Human Engine Labs

Scroll
Section

The Challenge

A de-esser set by threshold is only correct for one performance. The moment the vocalist leans in or backs off, the detector is wrong, and the engineer is back in the settings instead of listening.

Sibilance and the vowels that surround it share a frequency band. Tools that act on frequency alone dull the whole top of the vocal to remove one harsh consonant.

Transparency is the industry's most repeated claim and its least measured one. Buyers are asked to trust adjectives where they should be reading figures.

Section

Our Approach

01

We made detection phoneme-aware rather than level-triggered. The plugin classifies the consonant first and only touches the harsh ess, so the vocal stays open and the setting survives a change in delivery.

02

We treated transparency as something to measure, not assert. Collateral change to the non-sibilant band is published as a number — ±0.04 dB, two-sided gated — so it can be checked rather than believed.

03

We built the validation suite as a shipped artefact, not an internal gate: 146 checks published alongside the product, benchmarked against FabFilter Pro-DS.

04

We licensed the free tier for commercial work. The protocol is shared openly at every layer, and an instrument someone cannot use on a paid session is not shared openly.

Section

Deliverables & Outcomes

What we delivered

  • Sibilance — phoneme-aware, level-independent de-esser
  • Thunder Clarity — three-fader vocal restoration, non-destructive
  • VST3, AU, CLAP, and standalone builds
  • Free tier licensed for commercial work
  • Published validation suite covering 146 checks
  • Perpetual licences, never a subscription

Measured results

±0.04 dB

Transparency

measured collateral on the non-sibilant band, two-sided gated

5.1 ms

Latency

Studio mode, host-compensated; about 1 ms in Live

146 checks

Validation

published with the product, benchmarked against FabFilter Pro-DS

What happened next

Thunder Clarity extends the series from correction into restoration, pulling reverb, room noise, and bleed off a vocal on three faders.

Interested?

See how the protocol works

Each case study demonstrates how Respectful Mediation, Structured Cultivation, and Natural Harmony operate at scale.