Noise reduction on singing: Clean against Clarity Vx
Measured 8 November 2026 on a Mac, with the plug-ins rendering offline.
Waves Clarity Vx is one of the best-known neural noise reducers for voice. We ran it and our Clean module on the same noisy vocals and scored both with the same test we use for our own engines. Here is what we found, including where Clarity Vx does better.
The test
- Material: 90 seconds of two real vocal takes, with five kinds of noise added at two levels: 10 dB below the voice and as loud as the voice (0 dB). The noises are hiss, a fan with mains hum, clicks, traffic rumble, and "babble", a crowd of other voices.
- Scores: how much noise is removed between phrases; how close the voice stays to the clean recording while singing; and how much of the first 40 ms of each phrase survives.
- Settings: Clean in Studio mode with the neural engine at Amount 80; Clarity Vx at its preset Reduce Constant Noise and with its one knob at full strength (Voice). For Live mode, our zero-latency neural engine against Waves NS1, which is also zero-latency.
- A note on fairness: in our test host, every Waves plug-in went silent for about one second every 20 seconds. We left those four seconds out of everyone's score, so all contenders are judged on the same 86 seconds.
Results
Noise removed between phrases (median of the ten conditions, higher is more):
| removed between phrases | steady noise at 0 dB (hiss, fan, traffic) | crowd of voices | plug-in latency | |
|---|---|---|---|---|
| Clean, Studio, neural | 37 dB | 37–40 dB | 21–22 dB | about 40 ms of Clean (62 ms for the whole chain) |
| Clarity Vx, full strength | 32 dB | 47–70 dB | 0–1 dB | 43 ms |
| Clarity Vx, Reduce Constant Noise | 21 dB | 22 dB | 0–1 dB | 43 ms |
| Clean, Live, neural 0 ms | 24 dB | 24–28 dB | 12–13 dB | 1.3 ms |
| Waves NS1 (zero latency) | 6 dB | 6–20 dB | 0 dB | 0 ms |
Where Clarity Vx does better
- Very loud steady noise. With hiss, fan or traffic as loud as the voice, Clarity Vx at full strength removed 47–70 dB between phrases, against our 37–40 dB, and scored a little higher on our overall fidelity measure.
- Phrase starts. Clarity Vx keeps the first 40 ms of each phrase almost intact (about −1 dB). Our neural engine softens them by about 3 dB.
Where Clean does better
- Other voices. Clarity Vx treats a crowd behind the singer as voice and removes almost none of it. Clean removed 21–22 dB.
- The top of the voice. At full strength Clarity Vx removes almost everything above about 12 kHz, the breath and air of a vocal, and with loud hiss it also turns down 3–10 kHz. Clean stays within 3 dB of the clean voice in every band on the traffic test.
- At the same voice damage. At its preset, Clarity Vx changes the voice about as much as Clean does, and removes 21 dB between phrases against our 37 dB.
- Live. With zero latency, Clean removed 24 dB against NS1's 6 dB. NS1 keeps attacks better.
What this doesn't show
- The noises are synthetic and the vocals are dry pop takes. Clarity Vx is built for voices in real rooms, and has controls (and a Pro version) we didn't tune for this test.
- Scores are no substitute for listening. Clarity Vx's lost air may or may not matter on your voice.
- Clean in Studio mode is not a repair suite. For spectral editing, de-reverb or dialogue work, use dedicated tools.
Waves, Clarity Vx and NS1 are trademarks of Waves Audio Ltd. They are named only for comparison; OneMove is not affiliated with or endorsed by Waves. Measured 8 November 2026 with the version installed on that day. Our neural engine uses RNNoise (BSD-3, Xiph).