On-device real-time voice isolation for YouTube
ClarityIQ — Hear the voice, not the noise
ClarityIQ analyzes YouTube audio frame by frame — a deterministic statistical pipeline, not a heavyweight neural model — to estimate which parts of the signal are most likely speech, then boosts those while pulling back background music and steady noise. Everything runs locally as the video plays; no audio is ever uploaded anywhere to get the effect.
Analyzes the audio of the YouTube video you’re watching, frame by frame, to separate speech from background music and noise.
Running locally means inference is free and unmetered — ClarityIQ can reason continuously instead of rationing AI calls the way a cloud service must.
Who it's for
The kinds of moments ClarityIQ was actually built around.
Viewers of music-heavy commentary & vlogs
Following a creator's narration over a loud background track without reaching for the volume slider every few seconds.
Non-native and hard-of-hearing viewers
Understanding dialogue clearly on videos where background music or ambient noise makes speech hard to follow.
Commuters and earbud listeners
Watching or half-listening to a video in a noisy train or street without losing the narration under the soundtrack.
Podcast-style interview watchers
Cutting through an intro jingle or bed music on interview and podcast-style YouTube videos so voices stay up front.
Features
What ClarityIQ actually does, in depth.
Real-time speech/music separation
A deterministic statistical pipeline — spectral analysis, feature extraction, and an adaptive per-frequency mask — estimates speech probability and reshapes the mix live, with no round trip to a server.
Adjustable Voice Boost, Music Reduction & Noise Reduction
Independent sliders for how hard to push speech forward, how hard to pull back music, and how much steady background noise (hum, hiss, fan noise) to suppress — plus one Overall Intensity dial that blends the effect in or out.
One-click presets
Podcast & Interviews, Balanced, and Noisy Vlog / Outdoors starting points, still fully adjustable afterward.
Live before/after visualization
A real-time spectrum and scrolling spectrogram show exactly which frequencies are being reshaped as the video plays, plus a live speech-confidence readout.
Entirely on-device
No audio is uploaded to a server to produce the effect — analysis and enhancement both run locally as the video plays.
Pricing
One Gethen Intelligence account. Upgrade or downgrade any time.
- 5-day free trial of full voice isolation
- Live before/after visualization
- Unlimited real-time voice isolation on every YouTube video
- All presets & advanced controls
- Usage stats
ClarityIQ vs. the cloud-AI alternatives
A feature-by-feature look at how ClarityIQ compares to well-known tools in the same category.
| Product | Runs on-device | Works directly in the browser | No file upload required | Built for live YouTube playback |
|---|---|---|---|---|
| ClarityIQ | ✓ | ✓ | ✓ | ✓ |
| Adobe Podcast Enhance Speech | ✕ | ✓ | ✕ | ✕ |
| Krisp | ✓ | ✕ | ✓ | ✕ |
| NVIDIA Broadcast | ✓ | ✕ | ✓ | ✕ |
Comparisons reflect each product’s publicly marketed capabilities as of 2026 and may change as vendors update their offerings.
Frequently asked
Questions specific to ClarityIQ. See the catalog-wide FAQ for account and billing basics.
Does ClarityIQ upload video or audio to a server?
No. The speech/music separation pipeline runs entirely on-device as the video plays — no audio is ever uploaded to get the effect.
Does this actually remove the music, like a vocal-isolation tool?
No — the goal isn't perfect source separation. ClarityIQ estimates which parts of the signal are most likely speech and reshapes the balance toward it, which meaningfully improves clarity without trying to fully extract an isolated voice track.
What's the difference between Free and Pro?
Free includes a 5-day trial of the full real-time engine. Pro removes the time limit for unlimited use across every YouTube video.
Will this slow down or add lag to video playback?
The processing pipeline is designed for low latency — small buffered windows of audio, processed with lightweight statistical methods rather than a heavy neural network, targeting well under 100ms of added delay.
More from Protocol
One account, every extension.
PrivIQ
Catches sensitive data before it leaves the browser
ContextIQ
A privacy-first reading layer for any page
CartIQ
Price tracking that never gets tired of checking
Culinairy
Just the recipe — no life story, no ads
Aissociate
A research companion for investment banking, private equity, and credit analysts
StoreIQ
Your private AI employee for Shopify
A note on accuracy: ClarityIQ runs its AI reasoning on-device — nothing is sent to a server to get an answer. On-device AI is not guaranteed to be accurate, complete, or even correct — summaries, deal judgments, extracted information, substitution suggestions, memos, and detections can be wrong, incomplete, or missed entirely. Treat every AI-generated result as a starting point, not a final answer, and verify anything sensitive, financial, medical, legal, or otherwise high-stakes yourself before acting on it.