It’s two tones, though. Like G and the C below that. It’s a musical fifth. If you listen on headphones next to a keyboard you can figure which notes they are. They match some of the performed notes which is what’s so entertaining about this. You can’t just notch them out because two of the notes on the keyboard later will vanish.
My suggestion was to notch the hiss which is subtracted from the intro, not notch the track.
As I mentioned when the notched hiss is subtracted (using Noise Reduction on Audacity) there will be a faint tone audible in silent areas which is the same frequency as the notch, which can be removed using a gate. I have tried this and it works:the “two musical notes” are very clear and the hiss is eliminated, although to be fair my version is not as faithful as Stevethefiddle’s.
OK, if you’re not musical, or didn’t pay attention in music class, those are just the specific piano notes that happen to line up (mostly) with the two interfering tones at the top of that sample–in case that rings any bells with anybody. Having the two tones come out musical (as opposed to dissonant like a TouchTone® phone) is a problem, because the music is musical, too.
We warn people that two very similar sounds are difficult to separate from each other. This is in reference to the person who wanted to get rid of theater voices competing with a singing show. I think that person is stuck.
What happened when you changed the settings around a little? Change the 14dB suppression number to 18dB? Or higher? In each case, a higher number will give you less hiss.
The sample or profile at the first step is critical. Try getting the profile from a different part of the noise, or more or less noise. The theory is that you let the program “sniff” the noise only and then the program will try to subtract that sample from the show. We also warn people that if there are any voices in the sample, Audacity will try to suck all the voices out of the show.
Thanks for the new samples - one thing that I notice is that all of them are a lot louder on the right channel than on the left. This is one instance where the unusual behaviour in Audacity of its “Normalize” function can come in useful. The unusual thing about Normalize in Audacity, is that it normalises each channel independently, so it will correct the channel balance mis-match.
I put all the samples into a single track, set to 32 bit and Normalize to -0.3dB.
Select and duplicate from about 6 seconds to 11 second of the intro clip (for the noise sample.
Boost the treble frequencies of the noise sample gradually increasing from about 800Hz and rising to aa max of about +12dB at 2000Hz, then levelling off at +12dB.
Use noise reduction to grab this noise sample.
Run Noise removal with reduction set to 14dB, Frequency smoothing = 150, Attack/decay = 0.1