Hey guys, I want to compare 4 sounds to a given sound. Is there a way to find the closest matching sound out of those 4?

Best

Elden

Look at this external:
timbreID
http://williambrent.conflations.com/pages/research.html

It is a start?!

Looks promising! Thanks!
Don't get how it's working, though. ^^
It seems that it compares a given sound to an array of a bunch of sounds. That doesn't really help me... .
If I have 4 arrays of sound and want to compare them to a 5th sound to find out which of the 4 is nearest to the fifth; how do I do this with timbreID?

If I recall correctly, you record a couple of hits on a cup of coffee for example, then the external will tell you when you hit a cup of coffee (with probability returned).

Maybe it's not what you are looking for...
Cheers~

Maybe I could use timbreID, if I concatenate my 4 arrays to one array containing all 4 sounds. That would require timbreID to get me the index of the concatenated array for the section that contains the sound most matching the given sample. Can I get this index anyhow?

Regards

@elden Very good question... I didn't use the external. Only knew about it.
Sorry maybe you can poke the author...
Cheers~

Ok. Another approach to compare audio, then. Let's say I analyze the target sound for brilliance, centroid, fundamental frequency and other features over time and record the analyzis data in an array.
Now I derive a tolerance range from the analyzis data that other sounds have to fit in.
Of course all sounds are varying over time so they will at some points in time fit this range and at others they won't fit.
Now I measure how long they did fit and select the sound that fit the longest time. Could that be realized anyhow? How would you patch that?

cheers

@elden Probably be easier to use timbreID.
Just use 4 instances. Each is set to detect one of your 4 sounds. Whichever one triggers is the match you need.

Your last post basically describes rewriting timbreID and using it to match multiple sounds.

That will not work, because
1: I don't know the 4 sounds. I only know the 5th sound. The 4 others will be recorded live and
2: because the timbreIDs will all trigger at the same time. If you have one sound to be matched with another single sound, it will everytime itself be selected as most matching the target - no matter how different they are. Other suggestions?

Ok, I just saw, that TimbreID is giving at least the respective successive number of the training sounds you feed it with. That's a start! Now I have to have a mechanism that detects slow attacking onsets of sounds. Is there anything like that?