Poster SPOT THE ODD SONG OUT: an online game for collecting music similarity data
0 likes337 views
We presented this poster on DMRN+7. Similarity estimation is a key topic in Music Information Retrieval with many applications. In scenarios such as music recommendation, user expectations depend on music similarity. Perceived similarity is specific to the individual user and influenced by a number of factors such as cultural background, age and education. Our goal is to adapt similarity models to similarity data from users, taking into account cultural groups of users sharing common attributes (cultural indicators). At this point, there are few similarity datasets openly available, and none contains information on user background.
More Related Content
Featured (20)
Poster SPOT THE ODD SONG OUT: an online game for collecting music similarity data
- 1. SPOT THE ODD SONG OUT: an online game for collecting music similarity data 孫Daniel Wolff , 族Guillaume Bellec and 孫Tillman Weyde 孫City University London, Music Informatics Research Group , 族cole Nationale Sup辿rieure de Techniques Avanc辿es, Paris daniel.wolff.1@soi.city.ac.uk, guillaume-bellec@ensta-paristech.fr, t.e.weyde@soi.city.ac.uk Music Similarity and User Data Similarity estimation is a key topic in Music Information Retrieval with many applications. In scenarios such as music recommendation, user expectations depend on music similarity. Perceived similarity is specific to the individual user and influenced by a number of factors such as cultural background, age and education. Our goal is to adapt similarity models to similarity data from users, taking into account cultural groups of users sharing common attributes (cultural indicators). At this point, there are few similarity datasets openly available, and none contains information on user background. Collected Data The game collects relative similarity judgments of users on triplets of songs, where they are asked to choose one song as the "odd song out". If song B is chosen, then: sim (A,C) > sim (B,C) AND sim (A,C) > sim (B,A) Data is annotated with anonymised user attributes: Users can login with their acebook account. Given prior permission, user attributes are extracted from their profile data. Music Database More than 11000 clips in total, each min. 30 seconds long Datasets: 1.Million Song Dataset Subset Mostly Pop/Rock music, streamed by 7digital 2.MagnaTagATune Only classic genre subset used. Age group Music education Gender Listening habits Occupation (sector) Favourite genres Languages Music activities Current location (city) Favourite media items Birth location (city) Religious affinities Geoprofile of friends Political affinities The Game Interface Motivation and Rewards Sample Selection HTML5 web application runs on many devices w. browser Game look and feel Multiplayer Hop-on hop off: enter or leave anytime Odd One Out scenario 45 seconds time limit Progress of other players is shown Decision and timings are logged In-Match: Input-agreement of players is rewarded with points In-Game: High-score table motivates competition Points represent in-game currency: Buys new fancy avatars Buys music genre selection for triplet: Classical, Jazz, Country, Electronica, Latin, Rap, R&B, ... Social network: acebook High-scores can be shared with friends Game can be advertised to friends Result: Agreement for Each Triplet Overal High-score Towards a Generic Data Collection API Being extensible to multiple question types and scenarios, the CASIMIR API manages the storage of user input and user attributes, as well as the music presented and the statistical selection of samples. Additional data annotation tasks such as beat annotations can be easily integrated. Benefits of an Online API Current CASIMIR API Implementation CASIMIR API is provided by a PHP5.x server, via the industrial standard SOAP protocol. The stable open source implementation for Odd One Out music similarity annotations provides three API hooks: GetOSOTriplet(user_details,opt_genre): returns a triplet of clips with urls to audio. SendOSOVote(user_details): stores a vote for the Odd Song Out GetOSOAIVote(triplet): returns a AI vote for the given triplet A hook for beat annotations is currently tested. Existing system and implementation allows to focus on data collection application. - Build your own application using CASIMIR API! - The CASIMIR HTML5+PHP Game Framework HTML5 interface built on a open source GWAP framework for CASIMIR: Based on LimeJS and Google Closure - runs on any HTML5 browser Real-time multiplayer! Single player and pseudo multiplayer possible Modular architecture: Little implementation effort for new applications Create your own game or add your module to Spot the Odd Song Out! API defines annotation tasks independent of game or survey logic. Generic task & I/O definition Different (simultaneous) client applications Reusable Extendable Music, Annotations, User Info Database Defines generic annotation tasks and data I/O Selects samples Manages survey subjects CASIMIR API Client-specific code GUI, visualisation Game / survey logic Player, Login and Score management Application 1 Application2 Central collection database Real-time control on data sampling Oversight of concurrent surveys Possible aggregation of annotation data With unknown participant numbers, online surveys benefit from real-time sample selection CASIMIR API (see right) manages an expanding music clip subset for all API clients For Odd One Out similarity annotation task, minimal thresholds of connections between the comparison triplets are assured, before the subset is expanded. mi.soi.city.ac.uk/camir/game/ Connected Triplets Separate Triplets