Content Management for Electronic Music Distribution: What Are the Issues?

EMD (Electronic Music Distribution) refers to the transportation and distribution of digital music. The profusion of EMD systems spreading in the PC and Internet world makes it difficult to have a global understanding of the real issues at stake, to compare and assess EMD technologies and services. We present a survey of existing approaches by stressing on the nature of the underlying content management technologies. We show that there is a recurring compromise to achieve between two approaches: a hard way and an easy way. We illustrate these different approaches on real world EMD examples. The Dream of Electronic Music Distribution Although the representation of audio data in digital format has been devised a long time ago, the possibility to store and manipulate such representation with good sound quality for music titles is much more recent. The emergence of efficient audio compression technologies such as MP3 has brought about the possibility of easily transporting and broadcasting music data across networks: Napster had 80 millions registered users at its peak [3] and 1.76 million songs were downloaded in November 2000. Another consequence of the power of audio digitalization is that the granularity of music F. Pachet, Music Content Management for EMD: What are the Issues? 2 distribution has shifted from music albums to music titles. Electronic Music Distribution (EMD) usually refers to the technical issues of transporting music data across networks, copy protection and copyrights management. However, there is much more to EMD than telecommunication and protection. A major challenge of EMD is to allow the shift from a mass-market approach to a personalized distribution approach. Providing this digital link between music and people is not a trivial task for several reasons, and it still remains a dream more than reality. First, size. Estimations based on major label catalogues yield a total of 10 millions titles restricting ourselves to published, occidental, popular music. The number of Internet users is about 500 millions in 2000, according to a Nua survey. Traditional mass-market distribution consists in distributing only a small fraction of music titles (hits) to a large number of people: the fraction of so-called “active” titles in major label catalogues is estimated to about 1%. The EMD dream is primarily about proposing personalized distribution schemes that make more titles available to more people. Second, EMD touches upon our intimate relationship with music. Browsing music is different from browsing a traditional digital library: we don’t want to simply “access” or “find” music, as we would, e.g. for bibliographical references. Users do not always know how to specify what they look for (the language mismatch problem created by ontologies users do not understand, see [1]); nor do they always even know what they look for. The design of an EMD system requires therefore that we know more about what users want to do with music. However, EMD systems seem to abound, in a large number of incarnations: Digital Audio Broadcast, CD-on-demand, music downloading, music streaming, Internet radios, F. Pachet, Music Content Management for EMD: What are the Issues? 3 music file management systems, music servers including peer-to-peer communication systems, content-based music retrieval systems. From a technical point of view, these systems differ mainly in the nature of the inputs and outputs they connect together (e.g. servers to CDs for on-demand CDs, to amplifiers for Internet Radios, etc.). EMD should be able in principle to handle all kinds of music sources and destinations. In which respect do these systems achieve the EMD dream? For us the answer lies in content management: Only content management will provide an efficient link between listeners and music. We describe here the most important issues underlying music content management: from identification to content-based music search and retrieval and user interfaces. We exhibit an opposition between a “hard way” requiring brute force and sophisticated technology which provide objective information but are costly to develop and maintain, and the “easy way” approaches based on statistical analysis of superficial data, which are straightforward to implement but less reliable. Content Management: The Fundamental Technology of EMD Music Title Identification How can a system identify music titles? In the simplest case, identification information is added to the music data itself, for instance through ID tags in Mpeg files. The ISRC (International Standard Recording Code) was developed by ISO (ISO 3901) to identify sound and audio-visual recordings. ISRC is a unique identifier of each recording that makes up an album. Unfortunately it is not followed by all music production companies, and is hardly used in unofficial music sources, so the majority of existing digital music files do not contain any built-in identification. F. Pachet, Music Content Management for EMD: What are the Issues? 4 Worse, music data may not carry any external reference information: this is the case of Hertzian radio for instance. In this case, identification can be done either the hard way, by analyzing the signal, typically a portion of the music title, and matching it against a database of prerecorded music signals. This task is addressed by technologies such as Broadcast Data Systems (US) or MediaControl (Germany), and is used typically by copyrights management companies to infer radio play lists. The techniques used to perform the identification range from pattern matching to more elaborate statistical methods based on characterization of the evolution of spectral behaviors. The easy way consists in using external information on the titles when available. External information can be as simple as file names, with the difficulty that names are even less standardized: an artist such as “The Beatles” may be catalogued as “The Beatles”, “Beatles, The”, or any other combination. Other, more reliable external information can be exploited: The Emarker system exploits the geographical and temporal location of a user listening to a radio and requesting a song, and then queries a large database containing all radio stations programs. The approach is lighter, no signal processing is required, and can scale-up to recognize virtually any number of titles. Of course, it works only for titles played on official radio stations.