How to split a recording to music and talking parts

Question

I have recordings of a radio show that I'd like to listen to later. But I'd like to listen only to the music parts. Sometimes there are interviews or discussions without background musik I'd like to skip.

Is there a woy to do this automatically or at least split the audio into talking and no-talking segments?

Of course I can look at the track visually, and the non-music parts are easily visible, but scrolling through the whole recording is a chore, and with new episode every week this not a viable solution. A solution with minimal time/efford is required.

Answer 1

I was about to tell you what I was taught at university back in the early '00s, that de-mixing is still in experimental form and extremely hard but I thought I'd check if there's been any advances since and I found this SOS article. Unfortunately all the de-mixing examples on their site are 404. If you want automated de-mixing, they are your best bet.

I did find an example. You can watch it here.

Unfortunately, such a solution might end up removing any lyrics from the music itself.

Answer 2

Schizomorph seems to be talking about splitting when the music and speaking are at the same point. Everything he said is correct in that regard, but am I right that you mean there is music, THEN speaking, THEN back to music, with little to no overlap?

In that case, i recommend you do it the hard way. Get audacity and just highlight the vocal parts (drag along the top half of the track to highlight), and hit delete. It's really quick when you get into a groove. Then you can drag the clips closer together and export that as it's own file.

There is a way to automatically do it if there is enough difference in volume. Can't remember the plugin right now (remove noise I think), but you set a threshold where anything above it is kept, and below is deleted. I always run into problems with this, so I vote against it. If you are being efficient, you can do hours long clips in 10-15 minutes (well, I do at least)

Answer 3

You can do this with a Vamp plugin from the BBC called "Speech/Music Segmenter". (https://github.com/bbc/bbc-vamp-plugins/blob/master/README.md)
You will need a Vamp plugin host, Sonic Annotator should be fine for this.
(http://vamp-plugins.org/sonic-annotator/)
You can have Sonic Annotator write the timings to a .lab file which can then be imported into a label track in Audacity (you might need to change the file extension to .txt, but the format is correct.) It might also be compatible with other DAWs but I'm not really sure.