Sounds the same / same folder weird deletes

[meaning]

Hi @paultaylor,

I tested the same folder feature. I ran a delete duplicates task on my matched folders.

here is what I ran, FYI :

cmd_start = f"docker run --rm --name {container_name} -v /mnt/cache/appdata/songkong_matched_cleaning:/songkong -v /mnt/user/MURRAY/Music:/music {DOCKER_IMAGE_NAME} -d {os.path.join(DOCKER_FOLDER, relative_path)} -p "songkong_deleteduplicates1.properties"

where “songkong_deleteduplicates1.properties” is configured this way :

What I was expecting is that SK would go through my matched folders, till it will end in an album folder, then would delete tracks that indeed would be dupes, because they sound the same.

What happened is that songkong deleted tracks that are actually not even the same track. I send you a few reports for reference. You’ll see that SK deleted stuff like this (I just put something randomly so you can see) :

You can see songkong detected tracks as “sounding the same” that are not even the same track, nor sounds the same.

Might be good to investigate this as it is hurting libraries.

[paultaylor]

Hi, okay so it should find duplicates files if within same folder and same acoustic fingerprint. In the example you have given are in the same folder, and at the top of the report it says they have the same fingerprint

There are 3 songs with AQAAD0mUAEkSRZEGAAA… key

so if this is true then it is doing the correct thing.

Now, investigating further.

I received your report files, this contains the last 5 Delete Duplicates reports and I cannot see your example in any of these so which report is it ?

I looked at some of the other duplicates in the reports and they appeared correct.

They do seem like different songs but do seem to have same fingerprint so have you played the songs to check to see if they are different or not?

Probably there was no acoustid fingerprint for these songs before they were run in Delete Duplicates so Delete Duplicates generates the acoustic fingerprint on the fly for the purpose of finding duplicates. This can be done locally, but in order to get acoustid id we have to go back to acoustid server slowing things down somewhat so we just use the fingerprint. Because this is a long string we only show the first 20 characters so cannot see for sure they are same but checked code and should only be same if fingerprint is same.

[paultaylor]

Okay I have just looked at the release https://musicbrainz.org/release/18dce48b-343b-43ef-b7cb-25c0139fb718 all three songs are only 4 seconds long

This is the issue, the songs are too short to reliably create a unique fingerprint. I thought I had a minimum length limit in place but it seems to be missing

So I need to set a minimum length limit, but this issue is isolated to incorrect matches of very short songs.

[paultaylor]

Raised https://jthink.atlassian.net/browse/SONGKONG-2533

[paultaylor]

and fixed in

SongKong 10.2 Atomizer released 29th of December 2023

We are pleased to announce SongKong 10.2 Atomizer is released 29th of December 2023 [atomizer] This is a bug fix release with two bugs fixed for the Delete Duplicates task Bug SONGKONG-2533 In Delete Duplicates the Sounds the Same option can match songs that are too short to have reliable fingerprint SONGKONG-2536 SongKong doing nothing for some Duplicate If options when in Lite mode

[meaning]

Pretty damn solid Christmas support @paultaylor!