Help splitting several voices from a sound.

Hi there,

I am a very beginner with all sorts of sound processing. I recorded this sound with my digital camera containing my 3 months old son’s first loud laugh. :wink:

I could easily split the voice from the video. transfer it into a workable format that audacity can read.

now, I’d like to remove the grand-mother’s voice and all other noise that contains the sound.

I am guessing a feature that could automatically highlight frenquencies or voice tone and allow to cut/crop them would be awesome.

In the mean time, I am totally out of solution. anyone has any ideas how to proceed ?


To give an analogy, that’s like trying to remove a tree from a photograph and see what was behind it, but of course if you remove a tree from a photograph you don’t see what’s behind it, you just get an empty space. Similarly if you cut out the grand-mothers voice, you will not hear the baby’s voice behind it, you will get an empty space - silence.

<<<I am guessing a feature that could automatically highlight frenquencies or voice tone and allow to cut/crop them would be awesome. >>>

It would. But human voices tend to use the same frequencies even between men and women and children, etc. There is too much overlap to be able to separate individual people speaking.

For just one example, can you tell who is speaking the word “Sssss?” I can’t either. That sound is the same no matter who says it. There are a lot of words like that.

The more complicated tools like Noise Removal only work on sounds like air conditioning fans that stay constant over the whole show. Vocal Separation only works on stereo shows, etc.

There’s not a lot of hope.


thanks to you two.

I guess you guys are right and redo the recording will be much much easier :wink: hehe

I’d still like to see some kind of frenquency splitting/masking in Audicity, just for the fun of it

Frequency Splitting can be done with the Equalizer tool. Particularly in Audacity 1.3, you can save your work and generate two complimentary patterns, or just use one number for Low-Pass and High-Pass.

Frequency Masking is what Noise Removal does. Expose it to a Profile (sample of the evil sound) and it will generate a complex series of filters to (hopefully) filter out the evil and leave the good performance.

We had a performance posted a while ago which was textbook perfect for Noise Removal. It had one very minor irritating sound that was constant through the whole show. Noise Removal pulled the noise out of the show without affecting the quality almost at all.

Most people want noise removal to reveal a speaking voice underneath a thermo-nuclear event’s worth of sound garbage – or worse, what you want. Separate one voice from many.

This is one single piano note. G1, I think. Just up from the left side of the piano.

It’s either the first or second note in this short performance…

Exactly what are you going to filter out so the rest of the show survives?


Possibly easier to do using the High-pass filter and Low-pass filter.
Make a duplicate copy of a track (select the track then press Ctrl+D)
Apply a high-pass filter to one copy and a low-pass filter to the other.
For a sharp cross-over, use a high “Roll off (dB per octave)” setting.
Use the same frequency and “Roll off” setting in both filters.