Clear Voice LIVE

1. Welcome

Clear Voice Live is a real-time AI voice processor designed for live sound, broadcasting, conferences and streaming. It does four things at the same time, on a single insert per vocal channel, at zero output latency in ZERO mode (the default):

• Denoising — removes background noise, fans, HVAC, audience bleed
• Feedback prevention — kills mic-monitor coupling before resonance builds up
• Dereverb — tightens room reflections and hall tails
• Voice presence (called Debleed on the website) — keeps the spoken/sung voice itself untouched while pushing down bleed from drums, monitors and nearby instruments on the vocal mic

All processing is driven by an AI core that separates voice from background frame by frame. There are no notch filters to draw, no rings to chase, no static EQ.

The most important thing to know: where Clear Voice Live works best.

STRENGTH is the one fader that decides how the plug-in behaves. The colour scale next to the fader (also in the GUI overview below) marks the working ranges:

• 0 to 2.9 - very light processing, mainly useful for a quick before/after check.
• 3 to 5 - the normal working range. At 3 a typical setup gets a gentle denoise and soft feedback prevention. If you need more, move towards 5.
• 5 to 8 - strong intervention. Use it when the situation calls for it and listen carefully; the processing becomes audible.
• 8 to 10 - emergency range only: maximum suppression for a situation that would otherwise be unusable.

DEREVERB works only while STRENGTH is above 0. At Strength 0 the whole processing chain is bypassed, so Dereverb has no effect until you raise Strength.

2. Quick Start

1. Install via the supplied ClearVoiceLive-Setup-1.1.6.exe (Windows) or .pkg (macOS). v1.1 installs alongside an existing v1.0.0 — your old version stays exactly where it is.
2. Re-scan plug-ins in your DAW. The new version appears as Clear Voice Live 1.1 under the Dynamics / FX category. If you already own v1.0.0 you will now see two entries — that is intended.
3. Insert it on a single vocal channel — one instance per microphone (see Best Practices).
4. CPU: LOW. Latency: ZERO. Strength: 0. Dereverb: 0.
5. Increase Strength until the noise floor disappears without losing voice character. For the finest denoising on an important channel, switch CPU to HIGH — in feedback-critical monitor setups, stay on LOW, it is the most stable choice against coupling.
6. Bring up Dereverb to taste to dry up the room.

3. System Requirements

The plug-in is written so that the host can route mono or stereo channels freely. Stereo input is processed in stereo, with stereo-linked gain decisions to keep the image stable.

4. Audio Setup & Buffer Size

In ZERO mode (the default) Clear Voice Live's internal latency is zero and the plug-in itself adds nothing; in LOW latency mode it adds a fixed 127 samples (2.65 ms), reported to the host as PDC. The real latency you hear in a live performance — the so-called roundtrip latency — is determined by your audio interface, the host's buffer size, and the driver running underneath.

Roundtrip ≈ (driver IN) + (host buffer) + (plug-in chain) + (driver OUT)

Internal sample rate

The AI core inside Clear Voice Live runs at 48 kHz fixed for all hosts. When the host runs at 96 or 192 kHz the plug-in does internal sample-rate conversion before and after the analysis path. The output stays at the host's rate. This means:

• There is no audio-quality benefit to running the host at 96 or 192 kHz just for this plug-in — the AI work happens at 48 either way.
• There IS a latency benefit, because the audio driver's buffer is measured in samples, not milliseconds. A 64-sample buffer at 96 kHz is 0.67 ms; the same 64 samples at 48 kHz is 1.33 ms. Running the host at the highest rate the interface supports comfortably halves driver latency.

Recommended driver setup

Choosing the buffer size

Clear Voice Live is a live plug-in. Anything above ~64 samples introduces enough roundtrip latency that the artist can hear themselves twice through the in-ear monitor — once acoustically through the head, once electrically through the chain. That feels wrong and is unworkable for live use.

• 32 samples: tightest possible roundtrip. Use it if your interface and CPU support it without dropouts. Best feel for the artist.
• 64 samples: comfortable safety margin, still under ~3 ms roundtrip on a typical 96 kHz interface. This is the sweet spot for most live shows.

5. GUI Overview

The plug-in window is intentionally small and resizable. Everything you can change is on screen, no hidden tabs or sub-pages.

The plug-in interface — two faders, the key column (LINK, CPU, LATENCY, BYPASS, DELTA), three meters, the TRACK display and the PRESETS button.

Header

• Title plate — click anywhere on it to open the about overlay (legal info, license attributions, version).
• TRACK display — a segment display in the header centre showing which channel this instance sits on. It automatically abbreviates the track name your DAW provides (e.g. "BV1"); if the host doesn't pass a name, it shows the instance number. In the about overlay you can force it to always show the instance number instead.
• Renaming the TRACK display — right-click (Ctrl-click on Mac) or double-click the TRACK strip to give it your own name: type up to 4 characters (A-Z, 0-9, space), Enter confirms, Escape cancels. Leave it empty (or choose "Reset to DAW name" from the right-click menu) to return to the DAW track name. A name you set here wins over the DAW name and over "force instance number". It is saved with the session, not with presets.
• PRESETS button — opens the preset menu: factory presets plus your own preset files. See section 9.
• Status LED + label — green & "ACTIVE" when audio is processed, off when bypassed. During trial it shows "TRIAL Xd" or "TRIAL mm:ss" countdown.
• Version — current plug-in version, top right.

Link groups (LINK A / B / C)

The three LINK keys tie instances together across channels. Put every backing-vocal instance in group A, and moving a fader on one of them moves the same fader on all of them — relatively, like a trim: each instance keeps its own offset, the group follows the gesture. Buttons (CPU, LATENCY, BYPASS, DELTA) switch absolutely across the group. An instance can be in several groups; linking works between instances in the same DAW project.

Meters

• IN / OUT — input and output signal level, shown as RMS (average energy over the block), not momentary peak. RMS reflects the average level of the signal rather than its loudest individual samples.
• VOICE — the detected voice probability, 0–100 %. This is a confidence indicator, not a level — so its segments stay green across the whole range (no yellow/red), even at high percentages. A high reading just means "strong voice detected here."

Processing engine

All processing in this version runs on the CPU — no GPU is used or required, and there is nothing to configure. The AI model loads in the background the first time audio flows, so inserting instances and plug-in scans stay instant.

6. The Two Faders

STRENGTH (0 … 10)

The master amount of denoise / cleanup processing. Fader at 0 means the plug-in is effectively bypassed (gain = 1 across all bands). Fader at 10 means full processing.

Use this fader to dial in how aggressive Clear Voice Live is. The colour scale next to the fader in the GUI overview shows the working ranges: 3 to 5 is the normal range (at 3 a typical setup gets a gentle denoise and soft feedback prevention; if you need more, move towards 5), 5 to 8 is a strong intervention that becomes audible, and 8 to 10 is for emergencies only. Below 3 the plug-in does very little.

DEREVERB (0 … 10)

Dries up the room: reduces the reverb tail that follows each syllable and the reflections between words. New in 1.1, this fader drives two engines at once — a continuous tail suppressor, plus a continuous deepening of the AI model's own dereverb behaviour. At 0 the model runs in its neutral, reference-tuned state; raising the fader pushes it smoothly towards maximum dryness. At 10 the voice sounds tight and close-mic'd even in a genuinely reverberant room (at the very top a slight thinning of the voice is normal — that is the character of maximum dereverb, back it off a touch if you hear it).

Sessions and presets from v1.0.0 that used the SMOOTH fader still load correctly — the parameter lives on and keeps working, it simply no longer has a fader (the new model made manual smoothing unnecessary).

7. CPU & Latency Modes

CPU: LOW & HIGH

The CPU keys select how much processing power an instance may spend:

What the modes cost, relative to each other

Measured per instance in Reaper on a Windows laptop and on a MacBook with v1.1.0. Absolute numbers say more about the machine than about the plug-in, so what follows is the relationship between the modes — that part holds everywhere.

Two things to read out of this table. LOW in ZERO latency became cheaper than v1.0.0's SOFT mode (about a third less), so the mode you will run on most channels costs less than it used to. HIGH in ZERO latency became more expensive than v1.0.0's HARD mode — that is the deliberate price of the finer 256-band resolution that makes its denoising so clean.

Latency: ZERO & LOW

8. Bypass & Delta Buttons

Two safety buttons in the centre column. Both are locked by default to prevent accidental clicks during a live show. Each button has a small amber lock strip below it.

How the lock works

• Click on the lock strip below a button → toggles between Locked (amber) and Armed (dark).
• Click on the button itself → toggles the action, but only if the button is unlocked.
• The locked state persists across DAW sessions.

BYPASS

When ON, the plug-in is in audio bypass — the dry input passes straight through. Useful for rapid A/B comparison ("how much better is this with the plug-in on?").

DELTA

When ON, the plug-in outputs only the difference between the processed and the dry signal. You hear what the plug-in is removing — noise, room tail, feedback resonances.

Delta is a "nice to have" inspection tool. Never set values while in Delta mode. Delta lets you hear what's being removed, but it is not a tuning aid — the spectral character of the residual is not representative of how aggressive the processing is. Set values with the plug-in in normal mode, by ear, and use Delta only as a curiosity.

9. Presets

The PRESETS button in the header opens the preset menu. The factory presets come in two flavours — one submenu per latency mode, so every preset loads with the right latency setting for its job:

• Zero Latency — for live shows and monitoring (0.0 ms).
• Low Latency — the same starting points with the finer-resolution low-latency engine, for streaming/broadcast/recording.
• User — your own preset files, loaded from
  Documents\Clear Voice Live\Presets\ — presets you saved with v1.0 appear here automatically.

The factory presets

Saving and managing your own presets

• Save preset… stores the current settings (CPU mode, Strength, Dereverb, plus the legacy parameters) as a .cvlpreset file in the preset folder.
• Open presets folder opens that folder in Explorer/Finder.
• The file name is the preset name. Rename a file to rename the preset; copy files between machines to share presets. The menu re-scans the folder every time you open it — files you drop in appear immediately, no restart needed.

A .cvlpreset file is plain, readable text — you can view and edit it in any text editor:

# Clear Voice Live Preset
version=3
engine_type=0        (0 = LOW CPU, 1 = HIGH CPU)
latency_mode=0       (0 = ZERO, 1 = LOW)
strength=100         (0–100, the fader shows this ÷10)
dereverb_amount=0    (0–100, the fader shows this ÷10)

Preset files written by v1.0 (without the latency_mode line) load fine and default to ZERO latency.

10. Status, Trial & License

The header shows one of these states at all times:

Trial behaviour

A 10-day fully-functional trial begins on first plug-in instantiation. The trial is bound to the machine you started it on. Renewing the trial requires support assistance. Version 1.1.6 starts a fresh 10-day trial on every machine — also where an earlier 1.1.x trial has already expired — so you can re-evaluate the update with your own setup.

When the trial expires, the plug-in keeps loading and shows an "Expired" overlay, but audio is bypassed cleanly until a license is activated.

Activation

After purchase you receive a personal license file by e-mail — it carries your name and is cryptographically signed. Loading it is the whole "activation":

1. Open the plug-in and click the title plate (About overlay) → Load License…
2. Select the license file (or paste its text). Done — the header shows "ACTIVE" and the plug-in is licensed permanently.

There is no machine binding, no challenge/response and no internet connection required — ever. The same file works on your studio machine, your laptop and a rented FOH computer on show day (one license per person, usable on all your own machines, one at a time; see the EULA). New computer? Just load the same file again. Keep it with your backups.

11. Tips & Best Practices

1. Close the plug-in window during the show if you don't need it. The GUI (meters, animations) costs a little extra CPU; with the editor closed every instance runs in its leanest audio-only path. Keeping a window open to watch the meters is fine too.
2. Start with LOW in ZERO latency at Strength 0 and increase. For about 90 % of applications that is the final setting. If a channel deserves the last bit of finesse, switch it to HIGH — preferably in LOW latency, where HIGH costs the same as LOW; in ZERO latency HIGH costs several times what LOW does.
3. If soundcheck time permits, try both CPU modes per channel. Different voices respond differently to LOW vs HIGH. There's no universal best choice.
4. Feedback-critical monitor setup? Start on LOW. LOW is the most stable mode against acoustic coupling; switch to HIGH where sound quality is the priority and feedback headroom is comfortable.
5. Set the order: Strength → Dereverb. Dial in the noise reduction first, the room tail second.
6. Run a small Dereverb even on dry stages — it is feedback prevention. 1 – 3 on the Dereverb fader takes the room reflections out of the loop and helps the plug-in decouple a feedback-prone monitor setup; it costs nothing in sound.
7. Lock the buttons before the show. Both Bypass and Delta default to locked. If you've armed them during sound check, lock them again before doors open.
8. Never enable Delta on a live mic into a monitor or PA. See section 8 — Delta can cause feedback. Use it muted or in headphones only.
9. One instance per channel. Don't chain multiple instances of Clear Voice Live in series; the AI gating doesn't compound usefully and you'll spend CPU twice.
10. Insert first on the vocal channel, before compressors and EQs. The plug-in expects a relatively clean dynamic input. Heavy compression before the plug-in muddies the noise floor and makes the AI work harder. In particular, do not place it on a subgroup behind a compressor — the compressor constantly lifts the noise floor between phrases, which the AI then chases and you hear as pumping. Corrective EQ boosts (e.g. a low shelf) belong before the plug-in, sound-shaping EQ after it.
11. Close-mic sources only. Clear Voice Live is built for one voice on its own microphone (handheld, headset, lectern, lavalier). It is not suited for ambient, room or audience microphones — by design, everything that is not the voice gets pushed down, which on a room mic is the very content you want to keep.
12. Conferences & panel discussions: one instance per microphone always gives the highest quality — each channel gets its own AI. But a very good option for large panels: route all microphones through an auto-mixer (e.g. a Dugan-style mixer) into a single group, and place one instance of Clear Voice Live on the group. You save CPU, get consistent treatment across the table, and the auto-mixer keeps unused mics out of the path before they reach the plug-in.

12. Troubleshooting

The host doesn't show the plug-in at all

• Use the installer instead of copying files by hand. It puts the bundle in the right place and installs the required Microsoft runtime.
• If you must copy manually: copy the entire Clear Voice Live 1.1.vst3 folder to C:\Program Files\Common Files\VST3\ — that is the standard VST3 location every host scans. (On macOS the equivalents are /Library/Audio/Plug-Ins/VST3/ and /Library/Audio/Plug-Ins/Components/.) (...\Common Files\VST without the "3" is the old VST2 folder; most hosts also ignore %localappdata%-based VST3 paths.)
• v1.1 has no runtime dependencies — it is statically linked and self-contained. If you are still on v1.0 (or the host scanner blacklists the plug-in on Windows 10 LTSC systems such as the Waves LiveBox), install the Microsoft VC++ runtime once from aka.ms/vs/17/release/vc_redist.x64.exe (official Microsoft link), reboot, then Clear Caches and Rescan in the host. The Clear Voice Live installer also installs it automatically.
• Pro Tools: shows no VST3 plug-ins at all — it loads AAX only. Use a wrapper (Blue Cat PatchWork, DDMF Metaplugin) to run Clear Voice Live there.
• Apple Silicon / SuperRack Performer: on M-series Macs, Performer lists only ARM-native third-party VST3 and only from its current version with third-party VST3 support enabled. Run the host natively (not under Rosetta) and check that third-party VST3 support is switched on.
• Make sure your host is 64-bit and actually scans VST3 plug-ins.

The plug-in is bypassed and shows "UNSUPPORTED SAMPLE RATE"

Your DAW or interface is running at a sample rate other than 48 / 96 / 192 kHz. Switch your project to one of the supported rates and reload the plug-in. Most live setups use 48 or 96 kHz.

The plug-in is bypassed and shows "TRIAL EXPIRED"

Your 10-day trial has ended. Your audio passes through unprocessed (clean bypass — nothing is muted). Purchase a license at clearvoice.live and follow the activation steps in section 10.

The plug-in is bypassed and shows "TAMPERED"

The trial integrity check failed. Contact info@clearvoice.live for a reset and include the machine fingerprint visible in the About overlay.

I changed my machine / restored a system image — is my trial or license affected?

Your license is never machine-bound: the same license file activates on any computer you own. The trial re-binds itself automatically since v1.1 when it detects that the machine has changed (new system image, hardware swap) — the earlier “trial file does not match” message from v1.0 no longer occurs. If you ever see trial issues after a system change, write to info@clearvoice.live.

My host lists two Clear Voice Live entries — which one is which?

If you owned v1.0.0 before, that is expected and intended. The entry named Clear Voice Live 1.1 is the new version; the one named Clear Voice Live is your original v1.0.0. They are separate plug-ins for the host, so both can be inserted in the same session — useful for an A/B on one channel. The About overlay of each instance shows its exact version number if you ever need to be certain.

Updating from v1.0.0 — the two versions live side by side

v1.1.0 installs as its own product. It appears in your host as Clear Voice Live 1.1, right next to your existing Clear Voice Live v1.0.0, which is left completely untouched. Nothing is replaced, nothing is overwritten, and uninstalling either one does not affect the other.

• Existing sessions keep loading v1.0.0. A show file saved with v1.0.0 opens with the v1.0.0 plug-in — v1.1 never takes over on its own. To move a channel to the new engine, insert Clear Voice Live 1.1 and remove the old instance.
• Your presets and your license carry over. They live outside the plug-in bundle in Documents\Clear Voice Live\, and both versions read the same folder. Your own presets show up in v1.1's User menu with nothing to import. Your existing license unlocks v1.1 as well.
• Going back is trivial — the old version never left. Swap the insert back to Clear Voice Live and you have the v1.0.0 sound again, mid-show if you have to. There is nothing to reinstall.
• Re-check your gain staging when you swap a channel over. v1.1 no longer pulls the overall level down the way v1.0.0 did, so the same fader positions give you a hotter, more present signal.
• Still on the trial? v1.1 starts its own fresh 10-day trial, independent of where your v1.0.0 trial stands — evaluate the new engine properly even if the old window already ran out. v1.1.6 starts another fresh 10-day trial, even where an earlier v1.1 trial has expired. (An installed license unlocks all versions, of course.)

I hear crackles / dropouts

Check your CPU meter first — there are two very different causes.

1) CPU is high / at the limit. Raising the buffer size only hides this and defeats a zero-latency live plug-in — instead, free up CPU. Open the Windows Task Manager (Ctrl+Shift+Esc) or macOS Activity Monitor and close what competes for it:

• Background syncing (cloud-storage clients, OneDrive, Dropbox, iCloud)
• Other audio software running in parallel
• Antivirus or backup scans
• Browser tabs with heavy JavaScript / streaming
• Windows Update or system maintenance tasks

Close anything you don't strictly need during the show. If CPU is still borderline, switch some HIGH-mode channels to LOW.

2) CPU is low (a few percent) but you still hear crackles. Then it isn't processing power — it's real-time delivery: an audio, graphics or USB driver missing its deadline ("DPC latency"). The plug-in's own processing is glitch-free here (confirm with an offline render — it will be clean). Work through these in order:

• Power plan → High Performance (Windows) / disable low-power & App Nap modes (macOS). Laptops throttle the CPU on balanced plans — a common source of timing glitches.
• Check your drivers — a misbehaving background driver (graphics, Wi-Fi, USB) is the usual culprit. Updating your graphics and audio-interface drivers, and disabling Wi-Fi power-saving, clears most of these crackles.

The voice sounds nasal / hollow

Strength is too high. Reduce it until the voice character returns. If the noise floor still bleeds through at lower Strength, try switching the CPU mode from LOW to HIGH and re-dialling Strength — the full model reaches deeper at gentler settings.

I can't hear what Dereverb does

Dereverb requires Strength to be non-zero — it's part of the processing pipeline that Strength enables. Set Strength first, then bring up Dereverb.

13. Support

Questions, bug reports, license issues:
Email: info@clearvoice.live
Web: clearvoice.live

Open-source licenses for components used in this product are listed inside the plug-in's About overlay and at clearvoice.live/open-source.html.

End-User License Agreement: clearvoice.live/eula.html. Privacy Policy: clearvoice.live/privacy.html.