Removing background talking from recorded music clip?

[ahanysz]

...A guy on the Audacity forum ...

"Not in Audacity, no. You can recognize music because you're a human. To Audacity, those are all just sounds and it can't tell which ones are music. This is the same problem people have when they record a concert from the audience. Extra noises."

So yes, I don't see any software being able to recognize the difference between a cow mooing or someone talking or a horse galloping...

If a human can recognise the different layers, then your brain is running some algorithm. With modern advances in machine learning, software should be able to do the same thing. It's only a matter of time.

This is precisely what non-negative matrix factorisation is trying to achieve. Technical intro and a bit of a demo at Practical application in a small way: https://www.tracktion.com/products/novum Unfortunately I don't think there's an easy-to-use application yet which will apply the algorithm in a way that suits you. At this date it might even be easier to just transcribe the sheet music from repeated listening, then get someone else to play it for you. But never say never. The software you need might only be a year or two away.

[enroe]

At this date it might even be easier to just transcribe the sheet music from repeated listening, then get someone else to play it for you.

This is the best suggestion I could read here. So if you're really serious about it,
and want these pieces of music to be "freestanding," then that's the best course
of action.

If a human can recognise the different layers, then your brain is running some algorithm. With modern advances in machine learning, software should be able to do the same thing. It's only a matter of time.
...
The software you need might only be a year or two away.

Mmmm, you really are an optimist in God's kingdom. My estimation is that
this kind of separation will take another 50 to 100 years.

[ahanysz]

The software you need might only be a year or two away.

Mmmm, you really are an optimist in God's kingdom. My estimation is that
this kind of separation will take another 50 to 100 years.

You may be right. We'll know in another 26 years which of our guesses was closer :-)

[kidslow]

At this date it might even be easier to just transcribe the sheet music from repeated listening, then get someone else to play it for you.

This is the best suggestion I could read here. So if you're really serious about it,
and want these pieces of music to be "freestanding," then that's the best course
of action.

After spending a day with a spectral editor on the piece, I would agree.

However, Izotope RX spectral editor was actually OK with most of the talking parts. Just depends is it a loud single set of words (works great) or the background din of children playing (works not at all). It really doesn't work for long constant noises, like there was a portion where cart wheels or water wheel was turning underneath the section and Izotope had no answer for that. Waves Clarity Vx can deal with those types of noise but at the expense of adding mud and muffling the mix.

[mikehende]

Thanks a million for removing the noise for me Kidslow!

In my media player I use some TV themes such as Brady Bunch, Love American Style, Laverne and Shirley e.t.c as a jingle in between the regular tunes so I can use this clip in the same way.

BTW, the Little House series up to season 9 is available on a free App titled "freevee"

https://apps.apple.com/us/app/amazon-fr ... 1532153219

I watch little house on Roku Tv. I wish I could have recorded the clip to sound as exactly as it does on on the TV because it's different when played back on a stereo system but as mentioned "half a loaf is better than none"!

[lfm]

As kidslow mentioned a multi band expander/compressor.
- but you fokus to filter out the noise and talk
- then you put that on another track and invert phase to cancel out that
- so summed with adjusting level of extracted audio you might improve a little
- sometimes you render that noise track with a gate as well get get music part out as much as possible

That technique might be the closest you can get. But don't get your hopes up....