WARNING: This software uses web services provided by The Echo Nest, Last.FM and Music Brainz. If these web services have been changed or deprecated since jSongMiner was originally released in 2010, then these portions of jSongMiner may no longer function properly.

Overview

jSongMiner is a software package for auto-identifying songs and extracting metadata about them from various sources on the web and elsewhere. This software was originally designed for use in the context of digital libraries, but it can certainly be adopted for other purposes as well. For example, it could be used as a means of obtaining cultural features for automatic music classification, or even for annotating personal music collections. For those users who do wish to use jSongMiner in the context of a digital library, it has been designed specifically to be integrated with the Greenstone digital library software, and Greenstone modules have been implemented for using jSongMiner to build Greenstone collections.

jSongMiner begins by identifying unknown audio files using fingerprinting. Alternatively, it can also identify songs using metadata queries, either using metadata that is embedded in an audio file or using known metadata about a song. Once jSongMiner has identified a song, it can then extract metadata about the song from various sources, such as from The Echo Nest, Last.FM and Music Brainz web services, or from miscellaneous metadata embedded in the audio file. In addition to extracting metadata about songs, jSongMiner can also extract metadata about artists and albums associated with songs as distinct resources.

Once metadata has been extracted relating to a song, artist or album, this metadata can be saved as an ACE XML 1.1 Classifications file or as a return-delimited text file, or it can simply be printed to standard out. Each piece of metadata extracted by jSongMiner includes the metadata field label, the metadata value and the source from which the metadata was derived. Users can also opt to have the extracted metadata presented using unqualified or qualified Dublin Core tags, if desired.

In all, jSongMiner can extract well over 100 song, artist and album fields. Many of these fields can have multiple values (e.g. there may be multiple songs similar to a given song).

Like the rest of jMIR, jSongMiner is open-source and available for free. It is implemented in Java in order to maximize cross-platform utilization. The one exception to this is its use of the Echo Nest fingerprinting codegen binary, but the Echo Nest provides versions of this binary for use with Windows, OS X and Linux. In any case, jSongMiner can certainly be used without the local fingerprinting functionality offered by the Echo Nest codegen if necessary.

Advantages of jSongMiner

• jSongMiner provides an integrated and unified framework for acquiring many types of metadata from diverse sources of information about music. It keeps a record of resource identifiers in as many namespaces as possible while doing this, thus facilitating the integration of information from different sources.
• jSongMiner provides several different mechanisms for identifying audio files, so that if one approach fails another may work. jSongMiner also extracts some of the same metadata fields from different sources redundantly, so that if a given piece of information is missing or incorrect on one source, it may still be correctly found on another source.
• jSongMiner permits metadata extraction for both unidentified audio files and abstract pieces of music for which identifying information (e.g. title, artist and album) is known.
• jSongMiner keeps track of where each piece of metadata was extracted from.
• jSongMiner provides a command line and configurations file interface that provides users with a great deal of flexibility as to what they want extracted, from where it is to be extracted and how the extracted results are to be presented.
• jSongMiner is entirely open-source and free, and is as platform-independent as possible.
• jSongMiner has a highly extensible modular architecture that will facilitate the addition of additional sources of information to the jSongMiner framework in the future, as well as the incorporation of future improvements to existing sources.
• jSongMiner is designed to be easily integrated into other software, either through its very well-documented and accessible API, or through the use of its command line and configuration file interface. jSongMiner also allows known metadata about a song to be entered at the command line, so that information known to be true by users or other software can be easily incorporated into jSongMiner extraction queries.
• jSongMiner is designed specifically to be easily and usefully integrated into digital library frameworks, and has already been incorporated into the Greenstone digital library framework.
• jSongMiner provides the option of presenting extracted metadata in the form of unqualified and/or qualified Dublin Core fields. It includes an original Dublin Core schema carefully designed to suit the kinds of information that can be mined from the web and embedded metadata tags.
• jSongMiner is also designed specifically to be used in automatic music classification research. The integration of jSongMiner into the jMIR framework and its production of ACE XML files facilitate this. For example, one might use the song identifiers found by jSongMiner to extract classification features through the web, using software such as jWebMiner, or one might use the metadata extracted by jSongMiner itself directly as features.
• jSongMiner provides users with the option of saving extracted metadata as ACE XML or return-delimited test, and/or have it printed to standard out. These choices allow the user to choose the output format that is best suited to their needs (e.g. ACE XML for use in automatic classification, text for easy parsing at a later date or standard out for immediate access or post-processing by other software). jSongMiner also includes utility functions for use in translating and quickly interpreting previously saved metadata.
• jSongMiner allows the storage of metadata containing diverse character sets by providing the option of URL-encoding saved metadata.
• jSongMiner allows users to choose whether they want saved files about resources for which metadata has been extracted to be auto-named with content-derived unique identifiers, or whether command line-specified file names should be used.
• jSongMiner allows users to treat songs, artists and albums as separate resource types, and allows information to be extracted and saved independently for each of them, whilst at the same time maintaining information outlining the links between resources of the same and different types. Users also have the option of instead packaging artist and album metadata with song metadata, if they prefer.
• jSongMiner keeps logs of artists and albums for which metadata has already been extracted, thus allowing users to avoid repetitively reextracting the same metadata for artists and albums when a song is being processed that is associated with an artist or album for which metadata has already been extracted.
• jSongMiner allows simple access to sources of information on the web even if they must be accessed through a proxy server.

Manual

Many more details on jSongMiner are available in the jSongMiner Manual.

Related Publications

McKay, C., and D. Bainbridge. 2011. A musical web mining and audio feature extraction extension to the Greenstone digital library software. Proceedings of the International Society for Music Information Retrieval Conference. 459–464.

Questions and Comments

cory.mckay@mail.mcgill.ca

DOWNLOAD FROM SOURCEFORGE

-top of page-