An Experiment in the Automatic Creation of Music Which Has Specific Emotional Content

We present an experiment with HERMAN [9], a real-time music generator. HERMAN was designed to enrich the sensory environment of GhostWriter, an immersive educational tool. The aim of thus enhancing GhostWriter’s virtual environment, was to offer a more convincing sense of presence, to motivate and stimulate users. HERMAN supplies continuous, manipulable music. It has an input parameter which, it is claimed, determines the scariness of the music generated. Here, we evaluate HERMAN’s output; the results suggest a correlation between the input parameter and the scariness of the music as perceived by experimental subjects. 1 BACKGROUND 1.1 HERMAN and GhostWriter GhostWriter [9] is an computer-based educational tool, intended to support primary schoolchildren writing ghost stories. It presents them with a three-dimensional virtual haunted house. Children assume different identities within the environment and communicate via a textual interface. A teacher directs the exploration of the environment by cueing particular characters to speak, or by assuming an identity in the story. Text dialogue is stored during the interaction and, later, the children use it as a basis for their own stories [9]. This approach system is believed to encourage both creative writing skills and co-operation in the group. HERMAN is a real time computer-based music generator [3,14], developed to accompany GhostWriter. It has one input, the scariness level, varying between 0 and 99, which is adjusted via a graphical slider control. HERMAN’s music is intended to complement the visually stimulating 3-d haunted house environment. The combination gives the children a richer virtual environment to build on. The teacher can manipulate the scariness level and this dynamic input can be used to provide direction in the story. 1.2 The Implementation of HERMAN The HERMAN system is implemented in Java. The scariness parameter is passed through a mapping function, which converts its single value to a set of control parameters for the music generator. Using these control parameters to guide its rules, the generator creates intermediate music control data, which is rendered into MIDI format [10]. The detail of the system’s operation is explained by Stapleford [14] and de Quincey [3]. To provide some degree of continuity in the music, HERMAN stores phrases of music. These phrases may be repeated, identically or transformed to fit the current progression in the music. Phrases incompatible with the current scariness level cannot be used and new phrases are generated for this purpose [14]. The ordering of the phrases is constrained to one of eight forms, though the system is able to deviate slightly from these guidelines. Music is output on two MIDI channels, melody and accompaniment. A further planned channel for percussion was omitted due to time constraints on HERMANs development. 1.3 Defining Scariness To extract a notion of what creates scariness, HERMAN’s creators studied soundtracks from adults’ and children’s thrillers and horror films. Three people watched each film and independently made notes about “interesting structures” [3] evident in the music. A discussion about each film followed, to consolidate ideas. The films chosen were Alien [13], Vertigo [5], North by Northwest [6], The Princess Bride [8] and Beetlejuice [2]. The group concluded that these films covered a range of different scary styles, from horror and thriller through to comedy. Stapleford and de Quincey isolated several scariness-building aspects of film music, including: a loud theme followed by silence (Alien); heartbeat-like rhythms, with increasing speed or a sudden halt (Alien); disjointed, ‘bursty’ rhythms (Alien, Beetlejuice); fast sequencesof high-pitched strings (Alien, The Princess Bride); etc. Bernard Herrmann provided music for some of Hitchcock’s most suspenseful films, including Vertigo, North by Northwest and Psycho. When designing HERMAN, Stapleford and de Quincey studied Herrmann’s work in more detail, via the writing of Brown [1]. To create scariness (which is a kind of dramatic tension), Herrmann’s harmonic progressions break the rules that are supposed to govern Western tonal music [12], in particular, controlled ways. By not fulfilling a listener’s expectations [7] or by playing more or less discordant notes simultaneously, the music is able to deviate from the expected norm to something more uncomfortable. Texts on musical theory [12] were used to provide guidelines for the basic structure upon which tension rules were overlaid [14]. 2 AIMS 2.1 Verifying HERMAN Stapleford [14] and de Quincey [3] performed a preliminary evaluation of a small number of human subjects’ responses to HERMAN’s output, but it was not the primary focus of their work. They acknowledge that their method was flawed, and so the results, though suggestive, are not conclusive. We aim to investigate whether HERMAN’s output conveys a level of scariness dependent upon the scariness parameter, to the listener. Our work aimed to measure human response to HERMANs output in comparison with human-composed pieces of music, to give a more reliable measure of its ability to convey affective content. Using human-composed music in the experiment means that pieces with strong affective content for Western listeners may be introduced, and stringent comparisons may be made with them.