Web-based Instruction of Music Students through Internet Sources

We present a software tool that can help music tutors and students organize and search their musical digital library in an efficient way. Our system is based on fuzzy cmeans clustering and automatically organizes a collection of music files according to musical surface characteristics (e.g., zero crossings, short-time energy, etc.) and tempo. The system is complemented with semantic metadata, which allows the user to check the clustering result, correct existing textual meta-information (such as genre and album) and add additional semantic information (about lyrics, theme, other artists who perform similar music, musical instruments, etc). In this way, students are assisted in the organization of the domain information that they have learnt about in the past or wish to learn about in the future. Consequently, students are facilitated in finding more information on topics that they are interested in and consolidate the knowledge that they have learnt. Introduction The Internet provides many sources for music education and the Web has become an important mechanism for distributing music. Music students can benefit a lot from such sources as they can seek information about topics that they want to learn about. However, if students are left alone to surf on the Internet trying to find appropriate information, they may get lost in the Cyberspace without having the desired results. Many currently available music search engines rely on textual meta-information such as file names and ID3 tags as the retrieval mechanism. This textual description of audio information is subjective and does not make use of the musical content. Our aim is to provide a semi-automatic user interface that allows music students and tutors to interact with their music digital library in a flexible way based on combination of content and semantic information. In our research, we have addressed this problem with the creation of a software tool for a music digital library which is based on fuzzy cmeans clustering and automatically organizes a collection of music files according to musical surface characteristics (e.g., zero crossings, short-time energy, etc.) and tempo. To be able to search through a collection of music pieces and make observations about the similarity between objects that are not directly comparable, we must transform raw data at a certain level of information granularity. Information granules refer to a collection of data that contain only essential information. Such granulation allows more efficient processing for extracting features and computing numerical representations that characterize a music file. As a result, the large amount of detailed information of each song is reduced to a limited collection of features. Each feature is a vector of low dimensionality, which captures some aspects of a song and can be used to determine song similarity. Additionally, our system is complemented with semantic metadata which offers the user the ability to check the clustering result, correct existing textual meta-information (such as genre and album) and add additional semantic information. Semantic knowledge in our application includes knowledge about artist, composer, lyricist, genre, album or the collection in which a music file is included, title of a music file, keywords about impact on mood (e.g. relax, happy etc), keywords from lyrics and musical instruments. Moreover, our system is enhanced by a process of relevance feedback as a human-computer interaction technique which is used to build consistency and reliability between content-based and semantic information according to students’ and tutors’ knowledge and the relations produced by the system. As a result, future queries retrieve music files that users accept as more relevant to their queries. Thus, the music retrieval process is significantly improved as it combines objective type similarity computed from the acoustical signal with semantic information. The paper is organized as follows: Section2 describes the music metadata extracted from music files and used to manipulate and understand the audio content and semantic information. Section 3 describes the relevance feedback process, while Section 4 presents an overview of our system and its architecture. Conclusion and some directions for future research are given in Section 5. Meta-data for Audio information There are two kinds of meta-data for audio information, content independent meta-data and content dependent meta-data. Content independent meta-data are not directly concerned with audio content and could not necessarily be extracted from it. This information is subjective, textual and offers a semantic (high level) description of music files. The best known example of this meta-information is probably the ID3 tags, an extension to the popular mp3 format, which allows the user to add tags to the beginning or end of the file, containing such information as song title, artist, album name, genre,