Guitar Solos As Networks

GUITAR SOLOS AS NETWORKS Stefano Ferretti Department of Computer Science and Engineering, University of Bologna Mura A. Zamboni 7, I-40127 Bologna, Italy [email protected] ABSTRACT cerns music information retrieval, techniques worthy of men- tion are acoustic-based similarity measures [8], compression- This paper presents an approach to model melodies (and mu- based methods for the classification of pieces of music [9], sic pieces in general) as networks. Notes of a melody can be statistical analyses and artificial neural networks [10]. Finally, seen as nodes of a network that are connected whenever these in [11] artificial intelligence is employed to capture statisti- are played in sequence. This creates a directed graph. By cal proportions of music attributes, such as pitch, duration, using complex network theory, it is possible to extract some melodic and harmonic intervals, etc. main metrics, typical of networks, that characterize the piece. Using this framework, we provide an analysis on a set of gui- Studies on music can be based on symbolic data (music tar solos performed by main musicians. The results of this scores) or on audio recordings. Symbolic music data eases study indicate that this model can have an impact on mul- the analysis in several music application domains. For exam- timedia applications such as music classification, identifica- ple, finding the notes of a melody in an audio file can be a tion, and automatic music generation. difficult task, while with symbolic music, notes are the start- ing point for the analysis. Thus, in general traditional mu- Index Terms— Media Analysis, Musical Scores, Com- sicological concepts such as melodic and harmonic structure plex Networks are easier to investigate in the symbolic domain, and usually more successful [12]. 1. INTRODUCTION In this paper, we develop a model that allows capturing some essential features of a musical performance (a music Nowadays, there is a common trend in research to model ev- track). We will focus on melodies, and specifically on “so- erything as a network. In particular, complex network theory los”, which are a part of a song where a performer plays is a mathematical tool that connects the real world with theo- (quite often improvises) a melody with accompaniment from retical research, and is employed in many fields, ranging from the other instruments. It is quite common in music theory as- natural and physical sciences to social sciences and humani- serting that solos performed by musicians are bound to their ties [1]. Technological, biological, economic systems, disease technical and artistic skills. Indeed, musicians are recognized pathologies can be modeled in the same way. Focusing on for their own “style” in playing a solo over a music piece, that (multi)media contents, it has been proved that language, for identifies a sort of musical “language”, typical of that musi- instance, can be seen as a complex network [2, 3]. In this pa- cian. It is not by chance that an artist can be recognized from per, we show that musical pieces can be treated as complex others, and that we can classify artists in categories and hier- arXiv:1603.04979v1 [cs.SD] 16 Mar 2016 networks as well. archies. The goal of this work is to make a step further toward When dealing with audio, the main concern has been on the identification of the rules and characteristics of the music the issue of digitalizing it or to synthesize, represent, repro- style of a certain performer. If a music line is conceived of duce sounds, by employing a variety of sound generation as a complex network of musical units (notes, rests) and their techniques. Attention has been paid on transmitting, index- relations, it is expected to exhibit emergent properties due to ing, classifying, clustering, summarizing music [4, 5]. How- the interactions between such system elements. Complex net- ever, the idea of capturing some general characteristics of a works provide appropriate modeling for music as a complex melody (and harmony) is somehow an overlook aspect. In lit- system and powerful quantitative measures for capturing the erature, there are works in the field of computer science that essence of its complexity. focus on musical scores [6], as well as works on the auto- As a proof of concept, we retrieved and analyzed differ- matic transcription of the melody and harmony [7]. As con- ent solos of some main guitar players. Namely, the artists are Eric Clapton, David Gilmour, Jimi Hendrix, BB. King, This paper appears in the Proceedings of the 2016 IEEE International Conference on Multimedia and Expo (ICME 2016), IEEE, Seattle, 2016, pp. Eddie Van Halen. The selection of guitar as instrument and 1-6. these particular artists is motivated by the fact that there is a quite active community of guitar enthusiasts that share musi- We can represent a track as a directed network, whose cal scores on the Web. Scores are published and formatted, nodes are the notes played by the performer. When a per- usually, by employing description schemes that are alterna- former plays a note x, followed by a subsequent note y, we tive, easier and more intuitive to read, with respect to the add the two nodes x; y in the network and a directed link classic musical sheets. These schemes are based on guitar (x; y) from x to y. If, for instance, the player subsequently tablatures, and there is a wide list of software applications plays another note z followed by the note x, we add another and libraries to handle digital representations of such scores. node z and a link (z; x) leaving from z to the already exist- This simplified the creation of the database. ing node x. Networks can have cycles, i.e a performer can During the analysis, different measures are calculated, play two subsequent notes of the same type. Weights can be typical of complex network theory. We measure the length associated to links (x; y), depending on how many times that of solos, the dimensions of the networks, the degree distribu- link (x; y) is present in the network, i.e. how many times the tion, distance metrics, clustering coefficient and, finally, we performer plays a sequence of the two x; y notes in his solo. identify that the network representations of certain solos are small worlds. The paper discusses how these metrics are re- As concerns the amount of possible nodes in a network, lated to the “style” of the performer. in the occidental music, an octave (the interval between one The outcomes of such study can have an impact on mul- musical pitch and another with half or double its frequency, timedia applications and on studies of music classification which is the same note but lower or higher in pitch) is com- and identification, in general. While probably a music track posed of twelve sounds. Focusing on electric guitar, as an cannot be fully described via mathematical measurements, example, it is usually possible to create sounds belonging to nonetheless, these measures can help in discriminating among four octaves. (Actually, this is a simplified measure, just to the main features of a performer and a music track. Such give an idea on the amount of possible nodes). Then, each results can be employed as building blocks inside media ap- note has an associated duration. Two notes of the same pitch plications for the automatic generation of digital music with with different durations are considered as two different net- certain specific characteristics (e.g., the generation of a solo work nodes. Rests and chords are other possible nodes of a “a` la” Miles Davis). Such applications could be extensively network. In fact, it is quite usual to hear performers playing exploited in didactic scenarios, automatic music generation multiple simultaneous notes to create multiple voices in the applications, and multimedia entertainment. melody they are composing in the solo. Therefore, according The reminder of this paper is organized as follows. Sec- to this model a guitar solo (and, in general, a track) network tion 2 describes the network model for music solos. Section can be composed of hundreds of nodes. 3 discusses on an assessment on a list of guitar solos. Section Figure 1 depicts four examples of networks derived from 4 presents the obtained results. Finally, Section 5 provides four famous blues/rock guitar solos. It is interesting to ob- some concluding remarks. serve that these networks are quite different one from the other. Figure 1a shows a simple network, with some nodes 2. A NETWORK MODEL FOR MUSIC TRACKS that have higher in/out degrees (i.e., number of links entering or leaving a node). Figure 1b has a more linear structure, sug- Before dealing with the model, a brief terminology is intro- gesting that the melodic line was “simple”, with poor repe- duced to avoid potential ambiguities. A song is a musical titions of single notes. Figure 1c appears to be a clustered piece, which is composed of multiple, simultaneous sounds network, with some few nodes connecting the two clusters. played by different instruments. The part of the song played Finally, Figure 1d has a quite complex structure, with many by a single instrument is referred as a track. Notice that an nodes and with the presence of several hubs. This suggests instrument can play different tracks in the song (e.g. there are that it might be interesting to assess if different artists do multiple instruments of the same type, or the tracks have been have different characteristics that, statistically, produce dif- overdubbed). ferent types of networks. The solo is a part of a track where a performer is playing with unobtrusive accompaniment from the other instruments.

Load more