Clarity Vx cleans a vocal, it does not split a mix

The tool moment

Waves released Clarity Vx in March 2022, a plug-in the company's own product page describes as "AI-powered noise removal for vocals and voice," built to clean a voice track "for any mix, production, podcast or video." Unlike the multi-stem separators elsewhere in this batch, Clarity Vx does not split a mix into named instrument categories; Waves' own campaign page positions it primarily against dialogue and podcast cleanup, with music-vocal use as a compatible secondary application rather than the plugin's stated design target.

What the documents show

The plugin's own user guide states Clarity Vx "removes background noise (ambience) from vocal tracks and dialog" using "Waves Neural Networks that have been trained using millions of hours of examples to learn the differences between voices and noise," controlled mainly by a single knob and a choice of trained network models. That is Waves' own account of its training data's scale, not an independently audited figure, and the guide is explicit that the plugin operates on a single voice signal against its background, not on a full musical mix with drums, bass and other instruments to separate.

What stays with the musician

Because Clarity Vx cleans one signal rather than splitting a mix, an engineer using it on a vocal still has to judge how much ambience removal is appropriate before the voice starts to sound processed; the user guide itself warns that certain events, like an abrupt change in the noise profile or a sudden loud sound, "can cause the network to lose focus or compromise quality," and offers a manual reset rather than an automatic fix. That judgment about where clean tips into artificial remains the engineer's, track by track.

Judge it by listening

Waves' documentation describes the controls, a main knob, network selection, an analysis mode for stereo material, without publishing a listening-test comparison against untreated audio or another vendor's tool. This is an editorial suggestion: run the same vocal at the plugin's maximum removal setting and at a moderate setting, listening specifically for the "edges" of the voice the user guide says get removed first, then compare the trained network options the guide describes as biased toward either close-mic voice or secondary speech.

• At what removal setting does the vocal start to lose natural texture rather than just noise?
• Does switching between the plugin's trained network options change the result on your specific recording?
• Would a full multi-stem separator, rather than a single-voice cleaner, actually suit what you are trying to fix?

Clarity Vx is a narrower tool than a stem separator, and Waves' own documentation keeps that boundary clear even while its neural-network language echoes the marketing of full separation tools elsewhere in this field; the engineer still decides where noise reduction ends and vocal character begins.