Acon Digital rebuilt DeNoise 3 in Acoustica 8 with a machine-learning Automatic mode that removes the old requirement to capture a noise profile first. Manual mode did not disappear, and Acon still recommends it for complex or atypical noise that needs tighter control.
The Acoustica 8 restoration overhaul makes profile-free cleanup the obvious starting point, especially when you need to move quickly through dialogue, ambience, or music. Treating Automatic as the only serious mode would miss the more useful change, though. You can now choose speed first and fall back to a measured profile when the recording gives the model a harder problem.
The practical win appears on recordings with no convenient noise-only section. A voice memo may begin mid-sentence, a live recording may never drop to silence, and an ambience take can contain useful sound from start to finish. Automatic mode removes the old dependency on finding a representative patch of isolated noise before you can even begin.
You still need to listen critically rather than chase silence. DeNoise 3 includes a Noise Floor control that limits how much attenuation the processor can apply, so pushing the reduction ceiling lower changes how aggressively the background can be suppressed. Leaving a little stable noise can sound cleaner than stripping the floor until speech edges, cymbal tails, room tone, or low-level detail start behaving unnaturally.
Decades of spectral amplitude estimation research have treated noise suppression as a balance between reducing interference and limiting distortion to wanted speech. Your ears still make the final call. A quieter meter reading is not automatically a better restoration.
Profile quality matters. During learning, DeNoise is not reducing anything, so the section you feed it should represent the noise itself instead of a convenient chunk containing speech, music, or transients you care about. A contaminated profile can encourage the processor to treat wanted content as part of the problem.
The profile can also be saved and loaded. Saved profiles become useful when several files share the same recording chain, location, machine noise, or other repeatable background, provided the noise really remains comparable between them. Reusing a profile blindly across unrelated takes is a poor shortcut because the whole point of Manual mode is giving the processor a more specific description of what needs to go.
Manual mode adds controls that matter once artifacts start appearing. Soft Knee changes how gradually processing moves between noise and wanted signal, while Reaction Time sets how quickly the algorithm responds to level changes across frequency bands. Acon notes that longer reaction times can reduce tonal and burst-like processing artifacts, which gives you somewhere concrete to work when a hard cleanup starts sounding twitchy.
Mid/Side processing is available in both modes as well. When enabled, the stereo input is converted to Mid and Side before processing and decoded back to left and right afterward. It gives stereo material another route when the center and width carry different noise relationships, without forcing you to split the file into separate channels manually.
Automatic did not make Manual obsolete. Automatic is the efficient default when the unwanted bed is recognizable enough for the new model, while Manual is the controlled fallback when you can measure the noise more reliably than the processor can infer it. A difficult archival transfer, odd mechanical whine, layered environmental bed, or inconsistent recording chain can justify the extra setup.
Standard Edition owners face one more boundary. Full DeNoise 3 with its machine-learning Automatic mode belongs to Premium and Ultimate, while Standard includes DeNoise Light 3 and its profile-learning workflow, so the edition choice matters before any knob does if profile-free denoising is the feature you actually want.
The Acoustica 8 restoration overhaul makes profile-free cleanup the obvious starting point, especially when you need to move quickly through dialogue, ambience, or music. Treating Automatic as the only serious mode would miss the more useful change, though. You can now choose speed first and fall back to a measured profile when the recording gives the model a harder problem.
Automatic mode is the sensible first pass
Acon recommends starting in Automatic because it is usually faster and more effective on ordinary broadband noise. Hiss, camera noise, wind-like broadband contamination, and similar steady beds are the sort of material DeNoise is built to reduce, and the new mode can identify unwanted noise without asking you to find a clean pause first.The practical win appears on recordings with no convenient noise-only section. A voice memo may begin mid-sentence, a live recording may never drop to silence, and an ambience take can contain useful sound from start to finish. Automatic mode removes the old dependency on finding a representative patch of isolated noise before you can even begin.
You still need to listen critically rather than chase silence. DeNoise 3 includes a Noise Floor control that limits how much attenuation the processor can apply, so pushing the reduction ceiling lower changes how aggressively the background can be suppressed. Leaving a little stable noise can sound cleaner than stripping the floor until speech edges, cymbal tails, room tone, or low-level detail start behaving unnaturally.
Decades of spectral amplitude estimation research have treated noise suppression as a balance between reducing interference and limiting distortion to wanted speech. Your ears still make the final call. A quieter meter reading is not automatically a better restoration.
Manual mode earns its place on difficult noise
Manual mode works much closer to the DeNoise 2 workflow. You capture a noise profile from a section containing the unwanted sound, then DeNoise uses that measured profile while processing the material you actually want to keep. Acon positions the mode for complex or unusual noise sources rather than as the default path.Profile quality matters. During learning, DeNoise is not reducing anything, so the section you feed it should represent the noise itself instead of a convenient chunk containing speech, music, or transients you care about. A contaminated profile can encourage the processor to treat wanted content as part of the problem.
The profile can also be saved and loaded. Saved profiles become useful when several files share the same recording chain, location, machine noise, or other repeatable background, provided the noise really remains comparable between them. Reusing a profile blindly across unrelated takes is a poor shortcut because the whole point of Manual mode is giving the processor a more specific description of what needs to go.
Manual mode adds controls that matter once artifacts start appearing. Soft Knee changes how gradually processing moves between noise and wanted signal, while Reaction Time sets how quickly the algorithm responds to level changes across frequency bands. Acon notes that longer reaction times can reduce tonal and burst-like processing artifacts, which gives you somewhere concrete to work when a hard cleanup starts sounding twitchy.
Solo Noise exposes an overcooked repair quickly
Automatic and Manual share several controls, so switching modes does not mean relearning the whole processor. Solo Noise is one of the most useful because it lets you monitor what DeNoise is removing instead of judging only the cleaned output. Consonants, attacks, harmonics, or room cues becoming obvious in the removed signal is a warning to back off.Mid/Side processing is available in both modes as well. When enabled, the stereo input is converted to Mid and Side before processing and decoded back to left and right afterward. It gives stereo material another route when the center and width carry different noise relationships, without forcing you to split the file into separate channels manually.
Automatic did not make Manual obsolete. Automatic is the efficient default when the unwanted bed is recognizable enough for the new model, while Manual is the controlled fallback when you can measure the noise more reliably than the processor can infer it. A difficult archival transfer, odd mechanical whine, layered environmental bed, or inconsistent recording chain can justify the extra setup.
Standard Edition owners face one more boundary. Full DeNoise 3 with its machine-learning Automatic mode belongs to Premium and Ultimate, while Standard includes DeNoise Light 3 and its profile-learning workflow, so the edition choice matters before any knob does if profile-free denoising is the feature you actually want.
