Mercurial Tones Tether v1.2.2 [WiN]

• Product: Tether
• Developer: Mercurial Tones
• Version: 1.2.2
• Format: VST3
• Requirements: Windows 10 or later
• Source: shop.mercurialtones.com/products/tether

Tether is a leveling and gating plugin from Mercurial Tones built around one drawn line: wherever it sits on the loudness display becomes the output level, not a threshold chased with makeup gain. Audio above the line eases down, below it lifts up, tracked through Peak, RMS or K-weighted detection, and a floor beneath Target releases that pull so room noise stays put. AI Assist listens to a few seconds of audio and hands back starting Target, Floor and Pull values, rendered through Tether’s own engine, not a preset library.

Key Takeaway

Tether replaces threshold-and-ratio leveling with a single target line and a floor that stops quiet noise from riding along — useful for a mix engineer who wants a take to sit at a set loudness without manual gain automation. The tradeoff: total correction is capped at 24 dB, so material swinging wider than that still needs a rough pass first.

Drawing the Output Directly

Draw one line across a scrolling loudness view and that line becomes the output level — not a threshold with a makeup stage to chase afterward. A vocal take pulled to a Target of -16 LUFS lands there directly, without a second gain pass to compensate for whatever came before it in the chain.

Signal above the line eases down and signal below it lifts up, each by a fraction of its distance from the line rather than a fixed ratio. A verse sitting 6 dB under Target gets partial lift on the first pass and closes the rest of the gap as the pull keeps tracking it.

Peak, RMS and K-weighted detectors decide what “loudness” means for the pull, but the choice is wideband — a drum bus tracked on Peak clamps every hit the same way, with no separate detector for kick versus cymbal wash inside that one mode. Switching to RMS lets transients through and pulls the sustain instead, so source material dictates which detector fits.

The canvas draws two latency-aligned traces, raw level and processed result, so the amount of pull applied is visible rather than assumed from a single meter. An engineer riding a live-recorded acoustic guitar can see exactly which passage triggered the lift instead of reading it off fader automation after the fact.

Near, Far, and the 24 dB Ceiling

Low and High Pull run independently, so a vocal that dips quiet between phrases but only occasionally spikes loud can get strong lift on the low side and a lighter touch — or none at all — above 0 dB.

Focus biases where correction concentrates: Near tightens small drift, Far reserves the pull for big swings, Even spreads it across the whole range. A steady folk-guitar take benefits from Near; a live vocal with occasional shouted lines calls for Far, so quiet passages aren’t chased on every small dip.

The Natural Engine splits response into a macro/micro pair with crest-adaptive ballistics, so a fast transient and a slow swell get different speeds instead of one static time constant. Source modes for Allround, Drums, Bass and Vocal retune that split, but only the mode is exposed — the macro/micro split itself isn’t user-editable.

Parallel blend brings back the delayed dry signal for added weight without losing the raw transient, and total correction is capped at 24 dB with input and output trim at ±12 dB. A room mic sitting 20 dB under Target still closes that gap in one pass; anything wider than 24 dB needs a rough level pass first.

The Expander Hiding Inside the Leveler

Below the floor line, upward pull releases entirely, so a breath between vocal phrases or the room tone under a drum take doesn’t get lifted toward Target along with the rest of the signal. The floor sits up to 5 dB below Target with a 5 dB knee, so the release fades rather than switching off in one step.

In that region the floor behaves as a downward expander — quiet stays quiet instead of rising into audible hiss — running on the same detection path as the leveling itself rather than a separate plugin stage. A podcast vocal with a noisy room tone gets this without adding a second gate instance to the chain.

A 10 ms lookahead and a predictive gain guard stop a transient from sailing through a passage the floor just boosted, and the dry path is delayed to match so a parallel blend stays phase-aligned; the reported latency is compensated by the host automatically. The floor’s range is fixed at 5 dB below Target — a source with a wider gap between program level and noise floor than that can’t be separated cleanly with this control alone.

One Read, No Second Opinion

Pressing AI Assist and playing up to eight seconds of representative audio returns a signal-dependent subset of Target, Floor, Low and High Pull, Focus, Speed and Range, shaped by the source mode picked before listening starts. A drummer auditioning a room mic gets different suggested values than a vocalist running the same eight seconds through Vocal mode.

Candidate settings are rendered through Tether’s own detection engine rather than matched against a stored preset library, so the suggestion reflects what that specific signal measures rather than what a similar-sounding signal measured once. The listening pass takes as long as the audio played, with a Stop button available mid-listen and no separate analysis wait after.

AI Assist doesn’t keep adapting once the suggestion lands — it sets values once and stops, and current settings stay in place until the result is applied. It only ever proposes a starting point: it doesn’t distinguish between a take that genuinely needs heavy correction and one where the eight seconds sampled happened to be the loudest part of the performance.

FAQs