Announcement

Collapse
No announcement yet.

Different noise reduction technique

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Different noise reduction technique

    I was practicing the spectral editor and came up with this idea, which seems to work very well:

    With a tape recording, I isolated a very quiet portion of the tape (that is, not much going on in terms of actual signal).
    I copied that into its own file. I slowed it down to about 1/10th of its normal speed, and listened to it.
    I did spectra editing on any sounds I recognized as being actual signal (in this case music). I erased them from the file so that all I had was the noised at 1/10th of its normal speed. I listened several times to make sure there wasn't anything that sounded like music.
    Then I returned the file to its true speed, and listened again to make sure there wasn't any music. All you hear is noise.

    I then isolated the two channels of the original sound into separate files where right channel was only in the right channel, left channel only in the left channel. I took the noise file and duplicated it many times to match the length of the music file, and separated it so that there were two files, one with right channel noise in the left channel and one with the left channel noise in the right channel, and pasted those into the separated music files.

    That leaves me with 2 stereo files, one with the left channel signal in the left channel and the left channel noise in the right channel, and one with the opposite situation. I then used the file conversion filter to merge the two together, using (l+r or l-r) with the dB adjustment to get it set where you couldn't hear the noise at all.

    The result sounds very good. Sort of like an aggressive noise reduction, but with no artifacts introduced, and no loss in music because there's no music in the file, just the constant hiss, etc.

    Has anyone else tried this or am I just a little nuts, or is there an easier way to do this? The results are very good but it's a bit time consuming. I tried just "paste mix" where I inverted the phase of the noise, but then you have to guess without a preview, and the results weren't as good.

    Dan

    Sorry.. I left out the part where I took the noise and used a high-pass filter at around the spot where I could see there was a little bit of music, to eliminate most of the music to start with.
    Last edited by Craig Maier; 03-31-2019, 05:39 PM.
    Dan McDonald

  • #2
    And after all that work, I'm not sure I got any better results than just using CNF in artifact reduction mode.

    Ha ha...
    Dan
    Dan McDonald

    Comment


    • #3
      Hi Dan,

      Not sure, but I think that is difficult to beat the performance of the algorithms.

      Craig
      "Who put orange juice in my orange juice?" - - - William Claude Dukenfield

      Comment


      • #4
        Yes, it certainly seems like it. I'm always experimenting anyway, but I continue to be amazed.

        Dan
        Dan McDonald

        Comment


        • #5
          Dan,

          Did you then in essence, remove noise from the music by a straight subtraction of noise from signal ?

          Marc

          Comment


          • #6
            Yes - that was the idea. Just pull out the noise (no signal) and take it away. It works pretty well, but for all the hassle involved, I don't see it any better than the CNF. I thought it kind of fun to use the DSE and erase the signal, just leaving the noise, and you can do that so that you can hear really high signal by slowing the file down.
            Dan McDonald

            Comment


            • #7
              Hi Folks,

              If I recall my math correctly, truely random noise subtracted from truely random noise yields random noise. Of course, the noise on audio recordings is not truely random, but it is more random than coherent. That is part of the problem when doing things like this strictly in the time domain.

              Craig
              "Who put orange juice in my orange juice?" - - - William Claude Dukenfield

              Comment


              • #8
                Yes - I think that was going on. And of course it makes sense - I got good results eliminating constant high frequency noise, but things like hiss - I kept trying things like inverting the channels, adding, subtracting, etc. and it didn't do a great job there. I guess you get some noise reduction but you are also adding noise at the same time.
                Always good to think things through ahead of time.
                Dan McDonald

                Comment


                • #9
                  As I think about it, though, I wonder if creating the noise file would be a good way to give the CNF something to work with. Maybe I have too much time on my hands lately.
                  Dan
                  Dan McDonald

                  Comment


                  • #10
                    Hi -

                    Probably the best reference noise file is a dynamic one which changes as things move along - - - that is calculated automatically when one uses the Auto Spectrum feature in the CNF.

                    Craig
                    "Who put orange juice in my orange juice?" - - - William Claude Dukenfield

                    Comment

                    Working...
                    X