Matching Entities, a Generic Algorithm

In music or any data-related field, enriching data from an external source is a common still tedious task. External data can come from open datasets like wikipedia, web crawling (if legal) or any other data sources. The main idea is that you own locally your dataset, public or not, and you want to enrich it with public and rich data. In music, Musicbrainz and Discogs publish such datasets, that can be used for matching. As I recently worked with the RISM dataset (Répertoire International des Sources Musicale), I will use this as an example, to match person entities.

However, the algorithm I’ll build for you is source and field (and language) agnostic.

Translating track titles, part 2

In my last post, I dissected the construction of a classical music piece title, and laid out the foundation of an automatic translation. I ended that post talking about some exceptions and difficulties. One of them is about the concertos:

Translating track titles

In a previous article about track titles, I dissected the structure of a classical music track title, from the composer’s name to the movement title. I closed it article on an open question about translations:

About Release Dates

The Release Date of an album is a crucial metadata. Although it seems like a simple data, it is not, and is often misunderstood and misused.

Let’s have a closer look, and try to understand what is behind a Release Date of a music recording.

How to differentiate homonym artists

A common problem on platforms is that several artists can legitimately have the exact same name. In some other cases, questionable providers and artists will deliver a famous artist name to get a bigger attention from the platform audience.

These cases lead to one problem to resolve on the platform side: how to differentiate them. If they are not differentiated, the artist page and discography will be totally mixed up:

Classical Metadata: Track Titles

Classical music highly relies on works that are played over and over again. The oldest recordings are from the early 20th century (Enrico Caruso being the earliest famous recorded tenor).

A classical work doesn’t have a proper title. It’s title is thus created from several elements. Let’s dissect this full example:

Classical Metadata: the Artists Issue

Classical music is one of the genre which has been put aside from the global evolution of streaming music online.

Streaming music started of mainstream music, where each ‘song’ has one title and one artist. It was good enough for many genres but never fit classical music, for several reasons:

Using fingerprint to link tracks together

You have many reasons to link to different files representing the exact same song in a music catalog:

• You don’t want to hear it multiple times in a short amount of time
• You want to find one song when looking for it, not a collection of the same song
• You want to gather the related metadata for that same song (different languages, detailed artists…)
• You want to know the actual number of different songs in the catalog, with no duplicates
• You want to know who listened to song X, and not file X, file Y and file Z of the same song
• …

Music services at a glance

A bit lost in the current music services? I made this table up for you to see a bit clearer. Let me know if I made mistakes or if your favorite service is missing. I’ll update it soon.

 

Genres and tags

Genres are a major feature to sort and classify music. And when it comes to managing ~40M-track catalogs, sorting and classifying become a major concern.