Support Colour Coding Confidence

[steveyg777]

Sorry for the reply to your question being years later Paul, I’ve only just noticed you asked . I really find picards use of colour helpful to indicate how closely matched the song or release is compared to it’s original metadata. It makes seeing issues extremely quick and helps me navigate a plethora of songs more rapidly. Picards colour system is what’s kept me from abandoning it all these years.

To add further improvements i really think (correct me if songkong already has it) the ability to automatically apply changes if the confidence of the match is either 95%+ or as an alternative (and it being more like Picards colour system for similarity of new and original metadata) if the similarity score between new and original metadata is 95%+ similar (or whatever score /match confidence the user chooses). You could go even further as to allow this in preview mode too and the score can be adjusted and results reparsed on the fly so users can quickly see an overview of how accurate the matches/similarities are and then (semi automatically i suppose) authorise an auto apply. You could also allow even more granular choice by providing tick boxes next to releases and even individual songs so that you can add extra songs/releases or remove them from the auto apply (useful for the odd song or release that you KNOW is incorrect but 99 other releases/songs are perfect with your current confidence score adjustment).

I think this would revolutionise songkong in its control that it offers users while still allowing it to be an automated solution too.

[paultaylor]

Hi, Ive moved to its own topic.

So does Picards colour system represent the confidence of the match or simailarity between existing and old data because they are two different things. For example you may be matching an album that has no existing metadata, but it does have acoustids that link to mb recordings all from same album so you have confidence of a good match even though there is no actual match of metadata because there isnt any.

I happy to explorer this idea with you but I cant currently see the point in SongKong because there is a key difference between how Picard and SongKong match.

In the Picard system you can match using existing metadata with Lookup or you Scan individual fiels for Acoustid and it will shows albums that match the fingerprint. But the trouble is that these are separate tasks so that if your metadata is poor it may match to wrong album, and the fingerprints dont help, and with the scan it just looks at individual files rather than considering as a group so the songs can be split over multiple albumks.

In SongKong the songs are grouped (akin to clustering in Picard) and it uses a combination of exsting metadata, acoustid fingerprints, file/folder structure and Albunacks Ids (simailr to discids used by dbPoweramp when ripping a cd) to find potntial matches, these are then scored taking into account user preferences (e.g prefer Vinyl) and the best match is selected, but only if all songs in grouping could be matched to same album and we are confident it is a good match.

So there is no need to allow the user to decide the confidence because SongKong is only going to allow match if a good match. Now soemtimes it may match the right release group but not the preferred release not because the track listing is wrong but maybe you want the French version rather than the Spanish version, but these are minor differences that are best resolved by using Match to One Album and picking the release you want.

The other thing is SongKong is designed so that it can be run unattended, so any solution that require you to pick the right album as it goes along is going to be annoying. Selecting the right albums at the end is does not fit in with the SongKong model becausde this means that potentially a huge amount of songs would need to be saved at one point creating a File I/O bottleneck and if anything went wrong nothing would be saved.

Regarding control, we do have this issue Ability to lock certain fields on certain tracks so can never be overwritten that I am considering.

If you have some examples of SongKong matching the wrong album that would be helpful.

[steveyg777]

The colour system, I believe, is the similarity score between new and old metadata. However, maybe you could implement a toggle between similarity and the confidence of the match so you could use either which might be more insightful again in known whether to action the matches? In terms of songs with no metadata before, to me logically, it seems the similarity score would be very low because it’s completely different metedata?

“only if all songs in grouping could be matched to same album and we are confident it is a good match” perhaps this is what the colour system relates to best (?) - how does sk decide that it is a confident match? It is this confidence that could be represented with colour so that when a user is scanning through the results they can identify the general accuracy of the matches overall. If it helps to inform the user and increases workflow speed then i would say it is beneficial and a welcomed function.

It could potentially help users spot incorrect matches that they may otherwise miss.

Where you talk about “SongKong is designed so that it can be run unattended” i agree that it should be this way in an ideal world but this has always been the inherent problem with automated music tagging/organising. With a confidence score type function users could begin with 95%+ score matches or similarity scores and then move to lower confidence scores, assess the proposed results (aided by a colour system for quicker review) and then choose to action another batch of music, for example. And so on until it got to the point where the user felt they needed to intervene more manually.

Right now i don’t have any examples as I’m just getting started (again after a few years break from) with songkong. If i find any ill let you know.

I’m not sure if there is a visual workflow chart to help me see how each of sk’s functions work (including describing other features they can invoke such as filename masks or scripters, etc) but I’d love to be able to see each functions (fix songs, etc) process/system in place and options available for invocation;to help me get my head around songkong and it’s possibilities in detail. It would also help me identify how i can use the app creatively to achieve my goals with it. I talk about a chart of some kind because I’m more of a visual learner and a systems person.

[steveyg777]

Regarding " issue Ability to lock certain fields on certain tracks so can never be overwrittenthat I am considering" hasn’t SK already got this feature!? I thought it had?

I find one of the biggest challenges with a lot of taggers is when I need to tag and organise music that I’ve downloaded from where in the past where for example it’s a soundtrack compilation or an album but someone has obviously thrown music in from maybe the artist albums for those specific songs that’s part of this album that I’ve downloaded and when acoustic ID fingerprints it it revert it back to the individual albums and fragments the intended album into lots of different separate albums. I’m not quite sure why this happens if, for example, you can toggle the album on as a compilation but it’s very frustrating and causes me (and I’m sure other people) to have to revert to purely manual work and has the effect of being paranoid and suspicious about automated taggers; thinking that we can’t trust them for such albums. I’m guessing this is what the match all songs to one album feature is? I haven’t tried it yet but I’m guessing it’s that, is this possible to use in terms of the batch automation “fix songs" function?

[paultaylor]

@steveyg777

The colour system, I believe, is the similarity score between new and old metadata

Okay so we have a version of this really

When you run any tasks such as Fix Songs or Manual Edit, you can use one of the Browse menus such as Browse by Folder and then use the Updated Filter to only show items that have actually had their metadata modified

and then if you look at a song any chnages are show in in green wuith both old and new values shown

You can also use View As Spreadsheet so you can see all songs and any fields modified are highlighted in green

Regarding Confidence score, the concept dsoesnt make much sense to me. Consider this scenario you have two albums, one album has acoustic fingerprints for each song that match to exactly one recording that matches to exactly one album on MuscBrainz. The other album has no Acoustids linkled to anything on MusicBrainz but the metadata matches exactly withone album on Musicbrainz. So if I was going to try and calculate a confidence score the first album would score better than the second album, so if you set your confidence score limit to that boundary the second album would not match, so you are preventing SongKong making a match that it is happy with. Alternatively if we set the confidence to low does that mean we are then allowing matches that SongKong has red flagged and doesnt think are probably the right match.

I think a better approach is to let SongKong match the albums that it thinks are correct, because 99% are usually correct. Then you can use Match To One album or Undo Changes for any albums you are not happy with.

Also we recently added the inconsistencies section to the report , SongKong looks at the results and highlights anything that looked iffy, this also works on songs corrected by other applications such ass Picard, more details here

@steveyg777

I’m not sure if there is a visual workflow chart to help me see how each of sk’s functions work (including describing other features they can invoke such as filename masks or scripters, etc)

Check out this overview video

[paultaylor]

@steveyg777

Regarding " issue Ability to lock certain fields on certain tracks so can never be overwrittenthat I am considering" hasn’t SK already got this feature!? I thought it had?

No, not currently.

@steveyg777

I find one of the biggest challenges with a lot of taggers is when I need to tag and organise music that I’ve downloaded from where in the past where for example it’s a soundtrack compilation or an album but someone has obviously thrown music in from maybe the artist albums for those specific songs that’s part of this album that I’ve downloaded and when acoustic ID fingerprints it it revert it back to the individual albums and fragments the intended albumfunction?

So i think you are describing the issue with Picards scan feature that I described earlier, SongKong does not have this issue, I think you would find this video useful to watch. I reccomend you try out Fix Songs and then if you encounter any issues please run Create Support Files and share details.