Struggles / Questions AI Sound Enhancement

[John Ellenberger]

I had the misfortune of having to work with a video that had terrible echo in one of the two mics due to a PA system malfunction (they were mixed so no way to separate the tracks).  I am running PP 25.5 on a brand new fast Windows computer with NVIDIA.

1)  When I first get into PP it takes literally minutes before I am able to scrub on the timeline.  I am assuming this is the AI model chewing up my machine in the background but there is no indication why I am seeing these delays.  It mostly behaves normally after that but I occasionally get into a pause where the playback controls are just frozen.

2)  I can't figure out how to do a "reset" on the AI enhancement so that I can try a new mix of settings.  There are LOTS of sliders so it is hard to come-up with the right mix, particularly when some of them require major rerendering.  It's easy with the filter tools (just check the bypass box),  but the AI tool seems to place a number of controls in the Effects Control panel and its unclear how they interact with the controls in the "Essential Sound" panel.

3) As to the actual reverb elimination, it is doing a poor job.  Better than the DeReverb filter but the sound still has these little reverb artifacts where things are sounding normal, then all of sudden a couple of words will echo.   I have the eliminate reverb control cranked to the max and the mix control cranked to close to the max.

4) I see that a year ago people were saying that some "Web Based Podcast Version" was doing a better job.   Is that still the case?   I forgot where I can find this alternative tool(?) so a link / cross post would be appreciated if it is still worth trying.   Should I try the latest PP Beta?

[Peru Bob]

Try this:

https://podcast.adobe.com/en/enhance

[John Ellenberger · Accepted solution]

So to answer my own question: if you have a serious audio problem with a speech track, the web-based free tool is FAR superior to the AI "Sound Enhancement" feature,  even though its been over a year since this was first reported.  The process I used is as follows:

1.   Remove any "enhancements" that have been made to the affected tracks (right-click > Remove Attributes).  If your video and audio are still linked, be sure and uncheck the video attributes in the dialog box. I also delete the "tags" in the AI tool just to try and make sure it doesn't decide to go off on its own.
2.   Export the audio using Export > Media.  Turn the video toggle off and then set the format to AAC audio for the export.
3.   Upload it to the web tool, process, and then download the result file.  I just use the defaults settings.
4.   Import the result file and drop it in the sequence.  Mute the original audio track(s)

I waited until I had all my cutting done and then just processed the entire audio track.  I suppose you could work on individual clips before cutting but I haven't tried that.

As for some of the long rerendering pauses, I still see them with the AI tool more or less disabled.