Brain Science, Music Understanding, and Music Technology, The limits of current research and perspectives beyond.
The
Music Technology Group
of the
Universitat Pompeu Fabra, together with
the
S2S^2
and EmCAP European projects invite you
to a one-day workshop related to
multidisciplinary approaches to the understanding of music cognition,
limits of current research and perspectives at the interface between
music, technology, brain science and psychology.
The workshop will take place on November 30th 2005 at 16 pm
in the Auditorium (ground-floor) in the França building of the Universitat Pompeu Fabra
Passeig de Circumval·lació, 8
08003 Barcelona.
How to get there:
map
Any questions? Contact the organizers:
Fabien Gouyon
and
Hendrik Purwins.
Research in music cognition traces a back long way in the past addressing
fundamental questions at the core of human appreciation of music. For
instance, what is it that permits humans to compose, play and enjoy music?
What processes are involved in the communication of a musical emotion
between a composer and a listener?
Such questions have traditionally been tackled in different research
fields and via diverse methodologies. For instance, musicology formulates
theories about the syntactic structure of music and ways to notate it,
music psychology focuses on experience aspects of listeners by empirical
experiments, neurophysiology analyzes with empirical settings the
functional organization of cerebral structures when listening to music,
computational neuroscience seeks to model such structures (from single
neurons to larger brain areas) and music technology builds machines aiming
at performing tasks normally suited to humans, as composing or perceiving
specific aspects of music.
In this workshop, we aim at bringing forward recent achievements in these
different fields but most importantly highlight their current limits and
perspectives of pluridisciplinary research at their interface.
Experts in these fields will address questions as the following ones. What
insights can we expect from these research fields? Furthermore, what do
these fields fail to deliver at the moment? What are the hard unsolved
problems of each area? In which way can these fields take advantage of
each other? Do these areas converge, in the sense that they encircle the
problems from different sides, getting closer and closer to the point? If
yes, how fast do they converge? Is the aspect of conververgence sometimes
exagerated in public presentations of the areas?
This lecture explores the question 'can there be music without sound'? We will explore musical imagery, with some experiments, to demonstrate that even this covert behavior can be investigated experimentally. We will explain how experiments demonstrate that imagery seems to make use of the same brain regions as perception does. Cross modal associations will be treated, as well as BCI applications of musical imagery.
Part of the robustness of biological perceptual systems may be due to the ability of the brain to draw on a number of alternative processing strategies, and to flexibly adjust their relative importance appropriately for the current task and listening environment. In this talk I will consider the possible mechanisms underlying the formation of perceptual objects, and discuss how feedback inhibition and synaptic depression might be combined to achieve a flexible segmentation of the 'auditory scene'. One of the most prominent aspects of subcortical auditory processing is the formation of responses which are primarily phasic; this temporally sparse representation may be a necessary step in the segregation of perceptual objects, which are then integrated into a more sustained representation in cortex.
Aside from the main psychological and neuro-physiological approaches to musical mental activity there is an additional strategy of investigation which does not directly address psychological or neuro-physiological levels of description. This rather indirect strategy focusses on mathematical investigations of abstract musical structures and processes. These investigations typically depart from traditional music-theoretical observations and problems and attempt to relate musically prominent structures (e.g. the diatonic scale, or triads or seventh chords, certain successions of chords in musical pieces etc.) to prominent positions within a mathematical model. If this works fine, further investigation may then be devoted to the music-theoretical - or even a psychological - interpretion the mathematical model as such. I will illustrate this in the case of the "theory of wellformed scales" by Norman Carey and David Clampitt and add some notes on actual joint "work in progress" by Emmanuel Amiot and myself, which involves the application of finite Fourier transforms to those structures. Mathematics offers rich possibilities to reformulate theoretical ideas in different ways. This does - of course - not guarantee an easy success in search of connections between abstract music theory and brain science, but it seems worth to try.
In this talk I will sketch some of the inspirational ideas, facts and models that, being originated in fields such as musicology, cognition, information science or AI, have impinged the research and development carried out in the project "Semantic Interaction with Music Audio Contents". Some important shortcomings detected in those areas will also be commented in order to promote the debate.
This presentation compares three wide approaches to the pitch analysis of polyphonic music signals: (i) auditory-model based methods, (ii) signal-model based Bayesian inference methods, and (iii) unsupervised learning methods (such as independent component analysis). Instead of going into the details of each technique, their basic ideas and assumptions are discussed, and some results are shown. It is emphasized that all the mentioned methods can be seen to model the perception of music, yet they are based on very different assumptions and correspond to different aspects of human perception. Some are more "knowledge based" than others, and the chosen approach also affects the obtained analysis results. Recent results in the visualization of the pitch content of polyphonic music signals are shown.
Algorithms have been developed to perform tasks such as tempo induction, beat tracking, automatic transcription, genre recognition and melody (and bass line) extraction directly from audio signals. However, the "success" of these systems seems to be limited to 80 +- 10%, so considerable manual intervention is required to use these systems in practical settings. I argue that the bottleneck in computational systems is the encoding and use of high-level musical knowledge. To illustrate this claim, I present recent work on the audio alignment system MATCH, which provides an indirect method of extracting musical content from audio signals, by aligning a recording or a live stream with an annotated version of the same piece of music. MATCH uses an on-line time warping algorithm with linear time and space costs, allowing robust real-time visualisation of expression in musical performances of arbitrary length.