How Do Listeners Develop a Sense of Rhythm?

Introduction to rhythm cognition

A short introduction to music cognition, with a specific focus on the temporal aspects of music (an 11 minute video in Dutch with English subtitles, part of the UvA tv-series 'De Fascinatie' [The Fascination] directed by Bob van Gijzel).

Youtube options:

Link video 1: "Music is not just sound"

Link video 2: "Music is a human quality"

Link video 3: "A flawed listening machine is interesting"

Link video 4: "Music plays with the listener"

How do listeners develop a sense of rhythm?

Listeners seem to be able to develop a sense of rhythm from simply being exposed to musical examples. In other words, their sense of rhythm seems to emerge from merely listening to music. This hypothesis has been explored in studying several temporal aspects of musical rhythm such as rhythm, meter, timing, and tempo.

Rhythm, timing, and tempo

A performed rhythm can sound ‘mechanical’, ‘swinging’, ‘laid-back’, ‘rushing’, etc. This is caused by playing some notes somewhat shorter and others longer in duration. But how does a listener perceive the timing of these rhythms and recognize it as being ‘rushed’ or ‘swinging’? Why is a rhythm with a slightly shorter note not simply a different rhythm? How does one distinguish between timing and tempo variations in interacting with another musician?

Rhythm can be considered a combination several components (see Figure). The first component is the perceived rhythm, which can be represented on a discrete, symbolic scale. We will refer to this perceived rhythm as the rhythmic category, to distinguish it from the notion of performed rhythm, that is measured on a continuous scale. This is similar to rhythm as notated in a musical score (a symbolic notion), which is typically studied in music theory, and it is related to categorization as it is studied in music cognition. The second component is tempo: the impression of the speed —or changes thereof— of the performed pattern. Tempo is related to the notion of tactus as discussed in music theory, as well as the cognitive process of beat induction as studied in music cognition. The third component is expressive timing that describes the timing deviations in a performance (e.g., accentuating notes by delaying them a bit, or playing notes ‘behind the beat’), apparently independent of the tempo.

Emergent Meter

A quality of musical rhythm which differs from speech rhythm is meter. This is the perception of rhythm in terms of a hierarchy of repeating patterns, with varying levels of stress. We can test how strongly listeners develop a sense of meter by testing which rhythms they find more complex than others, and how complex they judge syncopated rhythms. If we use a very regular rhythm as a base, and distort the rhythm at one of its positions, we see which positions in a rhythm are 'more important' for the listener than others. Which positions serve as an anchor, changing the perception of a steady rhythm into a very complex one, and which ones are rather decorations or ornamentations, and can be omitted and still allowing the rhythm to appear steady and regular. For various stages in the human development different positions or a rhythm seem to have different importance. For example, we found that there are no crucial differences in highly trained musicians and non-musicians in their structuring of meter, and that both listener groups show a very regular and analytical structuring of rhythmic patterns. In newborns, we could as a first step show the perception of the most salient anchor-point in a rhythm - namely the very first position in a rhythm, termed the 'downbeat', showing that beat induction, the simplest case of meter, might actually be active very early in live (see illustration)


References

Role of Exposure

How does experience affect rhythm perception? Do listeners develop a finer sensitivity to rhythm from being musically trained (experience), or from simply being exposed to large number of pieces of music? To determine this, we conducted an online web experiment to determine whether exposure and/or expertise influences listeners judgements on expressive timing.

Click to view animation

As such, and in addition to what has recently been shown in, e.g., the pitch domain, the current study provides evidence in the temporal domain for the idea that some musical capabilities are acquired through mere exposure to music, and that these abilities are more likely enhanced by active listening (exposure) than by formal musical training (expertise), especially if people are asked to make judgments they are sure about, and try not to just guess. One interesting outcome of that study is, that listening apparently is not only shaped by major differences in musical traditions (e.g., different tuning systems all over the world, or the use of complex or simple rhythms), but that even within one musical culture, in this case the western music tradition, very subtle difference are found to be influencing the way people listen.

References

Expectation

A key feature of a good dancer or musician is their ability to anticipate future beats and synchronise with them. This requires the development of expectation. To study how human listeners develop a sense of expectation, one strategy is to build a computer model of our proposed theory of rhythmic expectation, and test it against musical examples and compare the models performance to human listeners.

We have proposed and implemented a computational model of rhythmic cognition that predicts expected times of musical notes. A dynamic representation of musical rhythm, termed a "multiresolution analysis" using an analytical technique called a "continuous wavelet transform" is used. This representation decomposes the temporal structure of a musical rhythm into frequency components, that may change over time. These rhythm frequencies are much lower than the frequencies of sound, typically in the range of 0.1 to 100Hz. Both expressive timing and temporal structure (score times) are represented in the multiresolution analysis and both contribute to determine the temporal expectancies.

Future expected times are determined from peaks in the accumulation of highly energised frequencies (termed "time-frequency ridges"). We evaluated this model using data sets of performed and generated musical rhythms, by its ability to produce expectancy profiles which correspond to profiles of the musical meter (time signature). The results show that rhythms of two different meters are able to be distinguished.

Such a representation indicates that a bottom-up process is able to reveal durations which match metrical structure from realistic musical examples. This then helps to clarify the role that expectations generated from memory must play, and hint at what it's contribution is to the formation of musical expectation.

A video of expectancy demonstration software made by Ricard Marxer Pinon, including the multiresolution expectancy model (Smith & Honing, 2008). The system synthesizes, as musical output, the expectancies generated in response to incoming musical sequences:

Click to view animation

A 20 minute (63Mb) screencast of the multiresolution expectancy model (Smith & Honing, 2008):

For more information:

Methodology

Web-based versus lab-based experiments

The study described under ‘Exposure’ resulted, quite unexpectedly, in an international discussion on Web-based versus lab-based experiments. In a series of commentaries we argued that, due to some important advances in technology, Web-based experiments have become a reliable source for empirical research. Next to becoming a serious alternative to a certain class of lab-based experiments, Web-based experiments were shown to potentially reach a much larger, more varied and intrinsically motivated participant pool. Nevertheless, an important challenge to Web-based experiments is to control for attention and to make sure that participants act as instructed; Interestingly, this is not essentially different from experiments that are performed in the laboratory.

References

Towards a measure of surprise

While the most common way of evaluating a computational model is by showing a good fit with the empirical data, recently the literature on theory testing and model selection criticizes the assumption that this is actually strong evidence for a model. This research explored the possibilities of developing a method selection technique that can serve as an alternative to a goodness-of-fit (GOF) measure. This alternative, a measure of surprise, is based on the common idea that a model gets more support from the correct prediction of an unlikely event than the correct prediction of something that was expected anyway.

References