Affective Calibration of Musical Feature Sets in an Emotionally Intelligent Music Composition System

Abstract

Affectively-driven algorithmic composition (AAC) is a rapidly growing field which exploits computer-aided composition in order to generate new music with particular emotional qualities or affective intentions. An AAC system was devised in order to generate a stimulus set covering 9 discrete sectors of a 2-dimensional emotion space by means of a 16 channel feed-forward artificial neural network. This system was used to generate a stimulus set of short pieces of music, which were rendered using a sampled piano timbre and evaluated by a group of experienced listeners who ascribed a two-dimensional valence-arousal co-ordinate to each stimulus. The underlying musical feature set, initially drawn from literature, was subsequently adjusted by amplifying or attenuating the quantity of each feature in order to maximize the spread of stimuli in the valence-arousal space before a second listener evaluation was conducted. This process was repeated a third time in order to maximize the spread of valence-arousal co-ordinates ascribed to the generated stimulus set in comparison to a spread taken from an existing pre-rated database of stimuli, demonstrating that this prototype AAC system is capable of creating short sequences of music with a slight improvement on the range of emotion found in a stimulus set comprised of real-world, traditionally composed musical excerpts.

Neural and physiological data from participants listening to affective music

“…During the calibration and training sessions recordings of pre-generated music were used as stimuli. These were generated with the music generator described in 20 and are included with the datasets made available with this paper. During the testing session the synthetic music was generated in real-time and online (the synthetic generator used to generate the music clips is described in 20 ).…”

“…These were generated with the music generator described in 20 and are included with the datasets made available with this paper. During the testing session the synthetic music was generated in real-time and online (the synthetic generator used to generate the music clips is described in 20 ). Because these music stimuli was generated in real time during the experiment they differed between participants and are not included in our datasets.…”

“…Because these music stimuli was generated in real time during the experiment they differed between participants and are not included in our datasets. Readers wishing to replicate this experiment may find the details of how to generate the music in 20 . We have also included the code used by this music generator and a short read-me file describing its use with the datasets.…”

“…Specifically, the BCMI was developed to allow users to increase the tempo of a piece of music by concentrating on kinaesthetic imagination of upper limb movement. Music was generated dynamically during the experiments via a music generator developed in the project and described in 20 . The tempo of the music could be reduced by relaxing and increased by kinaesthetic motor imagery.…”

The use of deep learning algorithm and digital media art in all-media intelligent electronic music system

“…For intelligent music system, scholars have carried out corresponding studies. Williams et al (2017), through designing the emotional-driven algorithm composition and applying 16-channel feedforward artificial neural network, found that the system could create short music sequences and effectively improve the emotional range in the stimulus set composed of real world and traditional music extracts [7]. Su et al (2017) proposed a new multimodal music recommendation system.…”