I'm trying to make two audio samples as similar as possible and my problem is that in one test-case their loudness isn't similar enough.
data1, sr1 = librosa.load(path_1, sr=None)
data2, sr2 = librosa.load(path_2, sr=None)
Here are my plots of data1 and data2 over time (s) - you can see data1 is a bit quieter at the beginning:
I thought I just have to normalize it, but then I realized librosa is already doing that on load.
So the problem seems to be, that the quieter parts are louder in data2 than in data1.
That sounds like a compression thing to me, so I tried to compress data1:
data1, sr1 = librosa.load(path_1, sr=None)
data1 = librosa.mu_compress(data1, quantize=False)
data2, sr2 = librosa.load(path_2, sr=None)
Unfortunately now data1's quiet parts are too loud.
So I tried to compress data2 too:
data1, sr1 = librosa.load(path_1, sr=None)
data1 = librosa.mu_compress(data1, quantize=False)
data2, sr2 = librosa.load(path_2, sr=None)
data2 = librosa.mu_compress(data2, quantize=False)
But then I'm back where I began...
I hope I made my problem clear ;)
How to compress the sounds the 'smart way' so that they both end up as similar as possible?
