Sun pm
From III: Human vs Machine performance in a beat induction
task.
Experimental: Spoken
N.P.McAngus Todd and S. Kohon.
Department of Psychology, University of Manchester, Manchester,
M13 9PL.
1. Background
Over the last few years we have been developing a general theory of rhythm and
timing which is consistent with the idea that the perception and production of
rhythmic sequences in music and speech is an elaboration of a more general
sensory-motor sequencing faculty of the left hemisphere. The theory is
implemented as a computational model in which an attempt is made to simulate the
principal brain structures involved in sequencing (Todd et al 1998, 1999). The
model takes sound samples as input and sychronises a simple dynamic system to
simulate beat induction.
2. Aims
The aim is to evaluate the model by comparison with the performance of human
subjects in a beat tracking task.
3. Method
In the first experiment 2 human subjects and the model were compared in a beat
tapping task for 160 samples of music (Todd and Kohon, submitted) with a variety
of rhythmic structures and tempi. The second experiment focused in more detail,
with a larger number of subjects, on a subset of the samples, a performance of
the 48 fugue subjects from the Well-Tempered Clavier by J.S. Bach. In all cases
subjects were instructed to tap the beat in synchrony with the music. The tapping
responses were evaluated by a number of parameters: including the main tapping
period, the time to reach a steady rate and the variability of the tapping
compared to the ideal beat.
4. Results
In the first experiment the model produced viable responses (responses with a
clear beat rate) in 70% of cases. The human subjects produced viable responses
in between 90 and 95% of cases. Of the viable responses, the beat rates of the
model corresponded significantly to the beat rates of the human subjects. In the
second experiment the model was indistinguisable from humans in it's beat rates.
The model also compared well with other comparison measures including time to
reach a stable beat and position of the down beat.
5. Conclusions
The results provide support for the model but also raise several issues for model
improvement not least that of scene analysis and the role of top-down processing
in the extraction of a beat.
Keywords: beat induction, computational model.
References
Todd, N.P.McAngus and Kohen, S. (submitted) Testing a sensory-motor theory
of rhythm perception: Human vs machine performance in a beat induction task.
Todd, N.P. McAngus, Lee, C.S. and O'Boyle, D.J. (1998). A sensory-motor theory of
rhythm and timing in music and speech. Proceedings of the International
Conference on Neural Information Processing. ICONIP'98. Japan, October, 1998.
Todd, N.P. McAngus, Lee, C.S. and O'Boyle, D.J. (1999). A sensory-motor theory of
rhythm, time perception and beat induction. J. New Music Research 23(1), 25-70.