CONFERENCE REPORT

CONFERENCE REPORT

AES 53rd International Conference

Semantic Audio26–29 January 2014

Barbican CentreLondon, UK

Semantic Audio is concerned with content-based management of digitalaudio recordings. The rapid evolution of digital audio technologies, e.g.,audio data compression and streaming, the availability of large audio

libraries online and offline, and recent developments in content-based audioretrieval have significantly changed the way digital audio is created, processed,and consumed. New audio content can be produced at a very low cost, whilealso large audio archives at libraries or record labels are opening to the public.The amount of available audio data grows greater each day, which is why con-tent-based management of audio recordings is essential to enable efficient useof these data. Aside from audio retrieval and recommendation techniques, thesemantics of audio signals are also becoming more important in object-ori-ented and intelligent production, including audio coding, editing, and pro-cessing. Recent product releases demonstrate this but many more innovativefunctionalities are becoming available.

Conference partners

268 J. Audio Eng. Soc., Vol. 62, No. 4, 2014 April

CONFERENCE REPORT

J. Audio Eng. Soc., Vol. 62, No. 4, 2014 April 269

Sponsored by

Conference cochairs Mark Sandler(top) and Karlheinz Brandenburg

London’s Barbican Centre, venue for the 53rd Conference

CONFERENCE REPORT

This was the motivation for the AES 53rd International Confer-ence, Semantic Audio, held in London, from January 26 to 29,2014. It was the second AES conference on the topic and follows

the highly successful 42nd Conference from 2011. The aim of the 53rdConference was to give a broad overview of the rapidly changing stateof the art, address many of the new scientific disciplines involved inthis still-emerging field, and continue to foster this line of interdisci-plinary research. From the outset, itwas agreed that the conference shouldfeature a large number of invited talks.This brought together the luminariesin the field and allowed all attendees toget a big-picture view of the state ofthe art in semantic audio and where itis heading.Although the conference was hosted

by Queen Mary University of London(QMUL), it was a joint effort of theUniversity and Fraunhofer IDMT. Itwas cochaired by Mark Sandler(QMUL) and Karlheinz Brandenburg(IDMT). Papers chairs were ChristianDittmar (IDMT), George Fazekas(QMUL), and Sebastian Ewart (QMUL).The local organization team was alsoled by George Fazekas.The technical program was full

and varied. There were 85 attendeesfor a program consisting of 17 tech-nical talks, including keynote andinvited talks, 34 poster and demon-stration presentations, two tutorials,and one panel session. The schedulewas arranged into sessions each deal-ing with particular aspects of seman-tic audio: music informatics andretrieval; semantic web, linked dataand ontologies; automatic musictranscription; semantic audio organ-ization and retrieval; intelligentaudio production; and speechprocessing and analysis.

TUTORIAL DAYThe first day was dedicated to tutori-als, and the venue was Queen MaryUniversity of London’s Media, Art andTechnology Lab. Chris Cannam and MarkPlumbley gave a tutorial on reusable soft-ware, which inspired researchers to pub-lish their code and data, and explained themeans by which this can be done. This wasfollowed by a tutorial overview given byMichael Terrell on the emerging field ofsemantic audio and music production.

MAIN PROGRAM – DAY 1The technical program began on the sec-ond day, and the venue for this was theprestigious Barbican, the largest performing arts center in Europe.Introductory remarks on the first day were given by Mark Sandlerand George Fazekas, and the program kicked off with a keynote

talk on music structure analysis by Meinard Müller. The technicalpapers this Monday morning set the scene for the conference, withoverview talks on creating research corpora and bridging the gapbetween audio and symbolic notations, as well as a thought-pro-voking talk on analysis of piano recordings that showed someimportant ways in which performers deviate from the musicalscore. An invited talk by Yves Raimond also generated a lot of

interest, especially from researchersinterested in “big data,” when hedescribed how a mixture of semanticweb technologies, machine-gener-ated annotations, and crowd-sourc-ing are exploited to enable the BBCto rapidly publish large archives ofmultimedia content. Perhaps the highlight of this day

was a live demonstration of theComplete Classical MusicCompanion by researchers from theAustrian Research Institute forArtificial Intelligence. They show-cased the first integrated prototypeof a system that identifies what pieceis being performed and at what posi-tion within the piece. It then tracksthe performance and will adjust tochanges by the performer. This wasdemonstrated live for acoustic piano,with volunteer pianists from theaudience. It was clear that thesystem could almost instantaneouslyadapt to whenever the performerswitched to a new piece or left outsegments within a piece. The proto-type was still on after the demon-stration had officially ended, and itprovoked laughter from the audi-ence when it tried to find the pianopiece that matched the programchair’s announcements aboutupcoming events.

MAIN PROGRAM – DAY 2The second day opened with akeynote talk by Gerhard Widmer,discussing some big-picture chal-

lenges in music semantics, and focusingon how a computer might identify, or evengenerate, expressivity in music. This wasfollowed by a session on music transcrip-tion, another big challenge, whichincluded technical talks on automaticallytranscribing bass guitar, detecting musicalkey (and using that to improve transcrip-tion), and unsupervised learning of thedown beats on drum tracks. The poster ses-sion that followed included descriptions ofseveral toolboxes, datasets and frame-works, as well as several presentations

related to intelligent audio effects and multitrack mixing.The last session of the day had interesting talks on frontier

research, one on modeling emotion in music, and the other on how

270 J. Audio Eng. Soc., Vol. 62, No. 4, 2014 April

Much enthusiasm was expressed during animated previewsof the various posters to be discussed in the poster session.

Yading Song, David Ronan, and Dawn Black ofthe conference committee during a break

CONFERENCE REPORT

to design audio-relatedexperiments as gamesto be played on socialnetworks or websites.This day ended with

the highlight of thesocial program, a galadinner on a boat cruisealong the ThamesRiver. The cruise wentpast many of London’smost famous sites,

i n c l u d i n gP a r l i a m e n t ,London Eye, St.Paul’s Cathedral,and TowerBridge. After thedinner awardswere given forbest paper (TomCollins et al.,“Bridging theAudio-SymbolicGap: The Dis-covery of Repeated Note Content DirectlyFrom Polyphonic Music Audio”), best presentation (KaterinaKosta et al., “PracticalImplications of DynamicMarkings in the Score: IsPiano Always Piano?”),best student papers(Christian Kirst et al.,“On-Line NMF-BasedStereo Up-Mixing ofSpeech Improves Per-ceived Reduction of Non-Stationary Noise”; and Daniel Gärtner et al., “Unsupervised Learningof the Downbeat in Drum Patterns”), best poster (ChrisoulaAlexandraki et al., “Using Computer Accompaniment to AssistNetworked Music Performance”), and best demonstration (AndreasArzt et al., “The Complete Classical Music Companion V0.9”). Theawards included products raffled out by Native Instruments andiZotope. A Reproducible Research Prize was awarded to Dan Stowell.

MAIN PROGRAM – DAY 3The last day of the conference began with a keynote talk by GaëlRichard. He presented an excellent overview of the state of the art inaudio source separation, focusing on informed techniques that offersuperior performance if some additional information is known. Thistalk was invaluable for those wanting to understand the rapidlychanging field of source separation, as well as to get a feel for some ofthe most important contributions in this large research area. This wasfollowed by an invited talk by Jay LeBoeuf, focusing on the challengesand opportunities for commercializing semantic audio research. Next,Pedro Pestana gave a stimulating talk summarizing a vast body ofwork he has performed on determining best practices for music pro-duction and how this may be exploited by intelligent systems. BothJay and Pedro were panellists in a special session immediately follow-ing this, where experts discussed the emerging field of intelligentaudio production. The technical program ended with a session onspeech processing and analysis, including a somewhat humorous, butstill rigorous, technical talk on the influence of alcohol intoxicationon speaker recognition, which used the “Alcohol Language Corpus”as a testbed. The conference ended on a high note, with a technicaltour of BBC’s new (and impressive) Broadcasting House.

J. Audio Eng. Soc., Vol. 62, No. 4, 2014 April 271

Carl Bussey with his poster onmetadata features forreverberation intensity

Tuomas Eerola and Simon Dixonenjoy a discussion during one ofthe poster sessions.

Delegates meet to share ideas around the demos and posters.

Committee members Mi Tian and Jordan Smith

Jordan Smith and Brecht De Man look atone of the demos together.

Presenters at the podium to deliver their papers

James Vegnuti and Mi Tian welcomedelegates at the registration desk.

Editor’s note: a USB drive or downloadable PDF of the conferencepapers can be purchased online at www.aes.org/publications/conf.cfm

CONFERENCE REPORT

272 J. Audio Eng. Soc., Vol. 62, No. 4, 2014 April

The assembled company gets together for a group photo at the end of the event.

GOOD TIMES AND IN-DEPTHLEARNING FOR ALL AT THE AES 53RD INTERNATIONAL

CONFERENCE

The band plays on after the water-borne gala dinner on the third night

The conference chairs enjoy congratulating awardees.

Committee with delegates: Brecht De Man, Katerina Kosta, MatthiasMauch, Shefki Kolozali, Panos Kudumakis, Carl Bussey, Yading Song

London’s Tower Bridge at night from the boat cruise

Meinard Müller and Olivier Lartillot

Katerina Kosta happy at havingwon the best presentation award