![](https://www.audiotechnology.com/wp-content/uploads/2023/03/HUSH_MAIN-pichi.jpg)
Review: Hush, Intelligent Noise Remover App for Mac
AI is here to rescue your noisy speech recordings.
With AI swiftly invading our everyday lives in the form of ChatGPT and AI-generated artwork, it was only a matter of time before AI spawned some genuinely useful audio tools.
Hush is most certainly one of them. It’s a noise removal Mac app for recorded speech that’s powered by machine learning.
Creator Ian Sampson kindly offered AT a trial version which I have been testing in anger with all manner of noise-riddled dialogue recordings. So far, Hush has exceeded my expectations.
The GUI is simple. A Mix control is the only real parameter impacting the results. Other than that, pick a destination folder and optional prefix and suffix for the output file, select the audio format, then drag ’n’ drop the audio into the window. Hush begins working straight away. It’s optimised for Macs with M1 and M2 chips running macOS 12 or later. The processing speed isn’t exactly lightning fast on my Intel iMac but the results are worth the wait. While the app accepts all types of audio formats I got the best results feeding it WAV files.
SPEECH FOCUSSED
Hush is clearly trained to zero in on human speech. I don’t pretend to understand how this works, but it’s evidently quite different to typical noise reduction plug-ins which use level and frequency content differences to isolate a source within a recording.
In one recording I put through Hush, there were two voices – the person on whom the lapel mic was clipped, and another person walking alongside him. There was a reasonable amount of background chatter and intermittent disruptions throughout. Naturally the second voice was softer and roomier than the first by virtue of the mic being closer to the first person’s mouth. At points, the second person’s voice all but blended with and disappeared into the background noise. Yet after being ‘Hushed’, both voices had been successfully isolated and presented with near equal clarity and presence, with the exception of a few brief moments. Both voices speaking simultaneously seemed to confuse Hush now and then, but generally Hush does a stellar job identifying and isolating the sound of a human voice.
![HushLightMode-pichi](https://www.audiotechnology.com/wp-content/uploads/2023/03/HushLightMode-pichi.jpg)
SQUEAKY CLEAN
I’ve done my fair share of vocal and speech noise reduction over the years. My experience is that even with the best tools of the trade, when used heavy handedly, it’s extremely difficult to avoid sonic artefacts like a loss of treble, weirdness in ‘s’ and ‘f’ sounds, and an overall wobbly, watery texture.
What impressed me about Hush is the way it preserves the natural timbre of the voice in the recording while cleanly, accurately and thoroughly removing excess noise and reverberations – this includes anything from steady hums and broadband noise to sporadic transient interruptions. Noisy stretches of audio are ‘gated’ to silence but phrases of speech start and end naturally. The result is quite like what you’d expect from a vocal booth recording, and thus takes extra processing – like compression, EQ, de-essing or enhancing – incredibly well.
It’d be nice to see more control in future versions of Hush over settings like sensitivity, background noise vs reverb attenuation, or even replacement ‘noise’ options to reintroduce a subtle ambience to a track that feels unnatural with the full anechoic treatment. I wonder if the program’s accuracy could also be enhanced if the user can specify whether it’s operating on a male or female voice, soft or loud, high or low, etc.
Ian plans to widen Hush’s application to support vocals and other musical sources in the future. For now, anyone who finds themselves frustratedly cleaning up noisy dialogue on a regular basis needs to check out the free trial and be instantly blown away.
Hush is one example of how AI is making what we once thought impossible incredibly accessible. It’s now available on the Mac App Store for an introductory price of US$49.99/AU$79.99/NZ$89.99.
Official press release below:
NEW AI POWERED NOISE REDUCTION APP FROM HUSH
Ian Sampson is proud to announce the release of Hush, a macOS app that uses AI to remove background noise and reverb from spoken audio. The app makes it easy to produce clean voiceovers, podcasts, audiobooks, and dialogue — without an isolation booth or professional studio.
Hush builds on the transformer architecture — the technology that powers generative models like ChatGPT and DALL·E 2 — to isolate speech with greater precision and fewer artefacts than previous approaches. It suppresses broadband noise, transient sounds like barking dogs or honking horns, and reflections from untreated rooms.
Hush is also one of the first professional audio tools to run on the Neural Engine, a subprocessor for machine learning on Apple Silicon Macs. The Neural Engine allows Hush to render audio quickly and efficiently, without heating up the computer or spinning up the fans.
Compatibility: Hush supports macOS 12 (Monterey) or macOS 13 (Ventura). An M1 or M2 Mac is recommended for optimal performance. The app does not require iLok or other third-party copy protection.
Price & Availability: Hush is available now on the Mac App Store for an introductory price of US$49.99. Sample audio and a 21-day free trial can be found at hushaudioapp.com.
Hush: www.hushaudioapp.com