Copy link to clipboard
Copied
I have been using AI Enhamcement for my voice recordings and it's amazing. Filters out all my background noises - a fan, the computer humming, other noises and leaves me with a pure voice. Like it was advertised. But i have two questions:
[A] the output has a percentage, and I can use the slider to change. But what it is a slidler OF exactly? It is labelled "90% Strength" as default, but strength - of what? Does it have more or less enhancement? More or less of the extraneous noises are removed? Is it the loudness of the resulting file? Something else?
[B] If i send it the voice channel from my multi-channel recording, the resulting enhanced recording has elements of the music that i thought had been elliminated by the multi-channel recording. My recording is being done on a Rode Rodecaster Pro desk, that is set up for multi channels - my mic is on channel 1, and the music is on channels 7 and 8. I thought that would enable the Rodecaster to separate the channels cleanly. Is there are reason other than crosstalk (which i should take to Rode for advice on) that might be the cause? I'm using Audition to separate the Rodecaster's recording into its separate channels - could that be the cause of the crosstalk? (In which case can i fix it in Audition?
(Audition is the latest update - my Creative CLoud is fully updated)
Cheers
Mike Kear
AFP Entertainment
Windsor, NSW, Australia
Copy link to clipboard
Copied
I think that 'strength' is a slightly misleading term, but I understand why they've used it...
For the AI to work, it has to separate what it understands a voice to be from everything else, and that it can do reasonably well. But because as a human, you can still distinguish the point at which the voice disappears into the background and as that sounds slightly artificial, there's a slider that effectively lets a bit of that background back, to 'mask' the join, as it were. If you use a short file, and experiment with the slider, it will become pretty apparent what's happening. The 'enhancement' level doesn't change, because that's what the AI has made of your voice - it's everything else. And that's why they've called it 'strength', if you see what I mean; it's the relative strength of each part of the processing.
If you can hear music on a vocal track that you thought wasn't there then it was probably buried in the noise, and now that the background is cleaner, it shows. There are two things to say about that; firstly that yes, it's almost bound to be crosstalk, and secondly it's there because for whatever reason, the AI process has interpreted it as something that might be speech - with the best will in the world, it isn't absolutely foolproof or perfect.