Hello.

I am looking for some advanced noise removal possibilities. I know that there are several audio editing software that require stereo track to get rid of noise but such method will often give you poor result. Now I want to get music from series that has 54 episodes. Music I want is played about 20 times in total. Sometimes partially. Is it possible that Audacity or any other software could take entire directory with 20+ tracks in it and then remove noise by comparing them to each other. As noise (speech, sound effects) is always different software could keep what’s common in tracks. Also some samples have only ending others only beginning but sometimes those partial tracks those that have clean music therefore they cant be left out. Program should understand wether given sample matches ending , middle, or beginning because if it compares ending with beginning it would output some garbage sound or even empty sound file because you cant find anything common by comparing beginning to ending. Therefore program should compare one track to 20 others to find out where it should be in timeline and then compare it millisecond by millisecond to all other tracks determine wether given part is noise or not.

Here’s example (x,y,z is noise):

axcdef
abydef
abczef

As you can see from this example noise be eliminated keeping char that is common in 2 of these alphabets like there is 2 ‘b’ but 1 x so x is noise. And you can also see that if we had only 2 lines that algorithm wouldn’t work anymore because we would have ‘x’ and ‘b’ both 1 therefore we cant decide which one to keep.

From that example you can see that the more examples you have the more complex noise you could eliminate. Imagine if you had 20 samples of a…z in example above. It would be possible to eliminate even 10 char long noise.

So the idea would be to automate cutting and pasting.

In my opinion it’s very technical and/or mathematical task and therefore should be easy for computers to do since the more samples the easier to decide if what you hear is noise or not.

Sound has 2 properties - frequency and amplitude. Computers can easily analyse both of these properties. Unfortunately these properties are not what defines the difference between music and noise. The same frequencies can occur in both noise and music - it is the context and combinations of sounds that distinguish noise from music. For example, a computer analysis of the sound made by someone saying “shh” will appear very similar to tape hiss. A human that is listening can easily distinguish the difference between tape hiss and a person saying “shh” because they understand the context of the sound. “Understanding” is not a strong suit for computers.