Feature music classifier

[dr4gn3l]

It is a draft resolving a part of #52 to classify music. If a name start with :

artist - Discography

• It will classify the torrent into music and try to found the artist with discogs api, if he founds it he add the genre into the database.
• It will also classify into music, torrents that have flac/mp3/wav/ogg in their name. But it will give some false positive. When a video was not found but contains an audio filetype it will classify into music.

I tried to keep the same philosphy as the video classifier. I'm not familiar with gorm so I add the source in the table metadata_sources by hand.

It currently works, I am using it why my crawler and the false positive are quite rare, it is almost always the fault of the mid filetype.

Regards :)

[mgdigital]

Thanks @dr4gn3l , from an initial glance this looks good! It's a non-trivial feature so will likely need some time and thought before we can get this merged but looks like the beginnings of the music classifier. Some initial thoughts:

• For the movie/TV classifiers I've adapted many regexes from the Radarr/Sonarr repos; the corresponding code for Lidarr is here: https://github.com/Lidarr/Lidarr/blob/develop/src/NzbDrone.Core/Parser/Parser.cs and in the enclosing directory (not that we have to copy all of these but it's a useful reference)
• It looks like the Discogs search is just finding artists, we possibly want to be finding either artists (for multi-release, e.g. discographies) or individual releases. Maybe the content reference would be something like or source: discogs_release/id: 1234 for individual release or source: discogs_artist/id: 5678 for multi-release (or if we know the artist but not the release)
• Looks like we're only saving some basic metadata about the artist and genres; what other metadata is available from the API and what might we want to integrate, e.g. dates, images, languages, summary paragraphs?
• You haven't yet done anything to analyse the quality, so codec, bitrate, source (CD/Web/vinyl rip) etc (this is what the preEnrich function is intended for) - I was thinking we'd add some audio-related fields to the torrent_contents table for this like we currently have for video. There are some useful pointers here: https://github.com/Lidarr/Lidarr/blob/develop/src/NzbDrone.Core/Parser/QualityParser.cs
• With reference to: if hasAudio := input.Torrent.HasFileType(model.FileTypeAudio); hasAudio.Valid && !hasAudio.Bool we might need to discriminate further here; e.g. .m4b is audio but would usually be an audiobook; .mid likewise wouldn't be a music release.
• Especially for discographies this might not play well with the dht_crawler.save_files_threshold setting which currently defaults to 50; need to either reconsider if this is the best value, perhaps we can just assume it's audio if it includes certain keywords, or perhaps the default needs to be increased
• We'll need to look at Discogs' policies and if we should embed an API key/push users to get their own API key as has been done with TMDB; what rate limiting will be needed?
• The other obvious metadata source for music would be https://musicbrainz.org/ - what are the relative advantages of Discogs/MusicBrainz and should we consider integrating either or both?
• "I'm not familiar with gorm so I add the source in the table metadata_sources by hand." - we'll probably have to add a small DB migration to add the metadata source, see the migrations folder
• Need to consider the WebUI integration
• Need to consider the Torznab API integration (though not necessarily as part of this PR)

[mgdigital]

BTW @dr4gn3l I've set up a Discord here that you're welcome to join: https://discord.gg/WecPrUFr

[mgdigital]

Closing this PR due to lack of response.