Lecture 6 Flashcards
Problem defining timbre
Timbre defined as everything except pitch, loudness, duration and
room acoustic influences. This is in fact a problematic definition: a)It’s a definition by negation: it does not pinpoint what timbre really
is. b) It does not include unpitched instruments as drums and
percussions, which surely have a timbre.
A problem is, that it seems hard to give a definition of timbre of
comparable compactness without stating it in this negative
fashion.
This definition and the apparent lack of better ones led to the point
that many people in music research think that timbre is a)
multidimensional, b) hard to define or understand, c) unimportant
in music.
I hope this lecture will show that a) is true, b) is not necessarily true,
c) is wrong!
Timbre concepts (9)
Multidimensional scaling - a data analysis technique for exploring multidimensional perceptual representations
Psychoacoustic correlations - the quantitative link between perceptual and acoustic dimensions
Spectral centroid - the center of gravity of a frequency spectrum
Attack time - the time from threshold to maximum energy at the beginning of a sound
Spectral flux - the evolution of the spectrum over the duration of a tone
Spectral deviation - the jaggedness of the spectral envelope
Acoustic specificity – an acoustic feature unique to certain
sounds
Individual differences in perceptual “weight” of a dimension
Klangfarbenmelodie – a musical progression ordered by timbres
Helmholz’s Klanfarbe?
In his seminal work published exactly 150 ago, Helmholtz laid out the first comprehensive theory of timbre. It was an early example of
interdisciplinary scholarship as Helmholtz managed to connect
diverse fields as physics, physiology, psychology and music.
Already Helmholtz thought about timbre (Klangfarbe) in a similar way in which the ANSI definition (above) did. However, in his work he
then also showed which components of musical sounds exactly
contribute to our sensation of timbre, which are important and
which are not.
His theory essentially culminates in the statement, that only the
number and relative strengths of partial tones are relevant for
timbre perception, phase is not.
Is that true? Well, in most cases, phase is in fact not overtly
important. But he forgot about the whole dimension of time! The
temporal envelope, the way in which sound’s onsets and offsets
and evolution are shaped is a main component of timbre (as we
will see later).
Note that all this was based on experiments with mechanical tools as tuning forks, sirens and resonators. Very crude tools, but cool work though!
Helm hilts essentials
The essential cornerstone of Helmholtz’s theory of timbre was the
focus on the structure of a sound’s partials. The importance of thisfor perception can e.g. be seen in formants.
Phase plays no major role for perception, sometimes however
contributes something we call roughness.
The major weakness of his account of timbre was the lack of the
temporal dimension.
Basic observations of timbre
- Dependant on spectrum
- Dependant on temporal envelope
3.
Timbre research 100 years later
About 100 years later, novel scientific methods were used to shed
novel light on the phenomenon of timbre. An important method
was MDS. By making use of this statistical tool, no verbalization of perception was required on the side of the listener.
The most important studies on timbre which make use of MDS were conducted from the 70s onwards by David Wessel, John Grey, Carol Krumhansl, and Stephen McAdams. The McAdams study is by far the most elaborate one, including a class structure for different groups of participant. As the main methodological principles as well as findings are similar over all these studies, we will focus the discussion on only this one.
The construction of timbre spaces
The basic approach of MDS studies. Given a set of sound stimuli,
listeners provide pairwise similarity judgements. These are
averaged over listeners in a similarity matrix which is the basis of the MDS algorithm. It then yields a geometric configuration which approximates the basic structure of the similarity data. E.g. If A is similar to B in average listener judgements, A and B will be close
in the obtained MDS space. The experimenter must then interpret the obtained dimensions of the space in order to connect
perceptual structure (as reflected in the space) and physical
structure (inherent to the sound stimuli). It is then concluded that
e.g. one dimension of timbre perception corresponds to the
rapidness of the attack of a sound.
Here is the 3dimensional space obtained by McAdams et al. (1995).
Green instruments are regular instruments of the orchestra, red
ones are hybrid instruments, e.g. produced by imposing a
temporal envelope of one sound onto a spectrum of another. The
space has three main dimensions. Note that for the third
dimension, the correlation between perceptual dimension and
physical descriptor, spectral flux, is only 0.54, compared to 0.94
with the tow other ones. This suggests that the interpretation of thethird dimension in terms of its underlying physical structure is not
as clear.
Note that in the original publications the labels “more-less” for
dimension 3 was mislabeled. They should be the other way
around.
Note also that along dimension 1, corresponding to the attack,
instruments mainly cluster in two groups (short vs. long). This
corresponds to the method of excitation of the instruments
(impulsive vs. continuous). This highlights the close connection oftimbre perception and source identification.
Spectral centroid dimension. The spectral centroid (in Hertz)
corresponds to the “center of gravity” of the spectrum. It will have low values, if the higher partials have low amplitude, it will have
high values if the higher partials are strong.
The third dimension of the timbre space cannot be interpreted as
easily as the other two before. The best fit in McAdams et al
(1995) was obtained for the spectral flux. It measures the
irregularity of the spectrum over time. The spectrum of a trombone constantly changes over time, in contrast to the spectrum of a
piano. Thus, the trombone will have higher spectral flux than the
piano.
In other studies as Krimphoff et al (1994) the MDS solution yielded a third dimension which could be interpreted as spectral deviation,
the jaggedness of the spectral envelope. Clarinets typically have
weak amplitudes on the even partials in low register.
The third dimension of the timbre space typically is harder to
interpret than the other two which always yield a dimension relatedto rapidity of attack and one to spectral envelope or centroid.
Timbre space and acoustic specificity
More advanced MDS algorithms can compute the degree of
specificity of each stimulus. This is a measure of how much the
stimulus is different from the others on a very specific perceptualdimension. E.g. the vibraphone is the only sound which has a
strong metallic flavor, this makes it very specific. It is thus likely
that listeners rely on these kinds of cues in their judgements.
Timbre space and individual differences
Not all listeners judge all sounds in the same way. This is accounted for by different classes of participants in the McAdams et al (1995)study. Note that the first two groups make similar judgements, but were reported to differ in their use of the rating scale. Class 3
focuses on the spectral dimension, class 4 on the attack and the
specificities. Important is that no systematic differences between
musicians and non-musicians were found (this differentiates
timbre studies from many studies investigating pitch).
Ex. Brain lesions
Samson et al. have studied the timbre spaces based on data from
patients with brain lesions due to surgery. They used a stimuli set
of size nine, generated by tones having three different numbers of
partials (1,4,8) combined with three different amplitude envelopes
varying in onset time. As expected, for normal subjects they
obtained a space which reflected these dimensions of spectrum
and attack time. The MDS solutions for the data by patients with
left temporal excisions still is relatively close to that of the normal
controls. However, a statistically significant difference is obtained
for the space coming out of the data from patients with right
temporal excisions. It seems that they do not incorporate the
dimension of temporal envelope at all, their judgments are solely
based on spectral differences. This provides evidence that the
right temporal lobe is important for timbre perception and particular for amplitude envelope information extraction.
Limitations
While MDS is an important technique for studying perceptually complex phenomena as timbre, there are several (inherent) limitations to most studies using it.
The spatial structure depends on the sounds that
are included in the testing
Use unison pitch only. Structure of space also valid for non-unisons?
Use one dynamic level only. Structure of space also valid for other dynamics?
Note that dynamics, pitch, and timbral descriptors (e.g. spectral centroid & flux) correlated! E.g. change in pitch implies shift of spectral centroid.
Timbre space summary
Most importantly, timbre is a multidimensional attribute. Its acousticalcorrelates are of spectral, temporal and spectro-temporal nature.
Timbre = multidimensional attribute!
Using MDS it was shown that both spectral (centroid) and temporal (attack time) contribute to how we perceive timbre
A third dimension of the space is in most studies related to the spectral envelope (spectral deviation vs. Flux)
Much future research remains to be done…
Timbre and source id?
MDS spaces often group (without any prior intention) sounds
according global source-excitation properties. For instance, a
vibraphone struck by a mallet will have higher spectral centroid
than a marimba struck by the same mallet (i.e. spectral centroid
reflects material properties). On the other hand, a bowed
vibraphone will have much lower attack time than a vibraphone
struck by a mallet (i.e. attack time reflects source-excitation).
However, the explanatory power of these acoustic descriptors are limited, as can be seen in the next slide.
MDS relate timbres by abstract acoustical properties (e.g. “Does this sound have a low vs. high spectral centroid?”)
But: timbre also the major tool for listeners to infer the identity of a sound source (e.g. “Is this a violin, a vibraphone, or a voice?”)
Both positions not mutually exclusive: timbre space structure reflects how sounds were produced (e.g. type of source excitation and material of source)
However, basic acoustic features don’t explain differences in recognition (Agus et al. 2012)
Compare processing speed (“go/no-go”) in reaction time study
Create acoustic “chimeras” with separate spectral and temporal features
I.e. neither spectral, nor temporal features alone can fully explain source recognition
Agus et al. looked at the speed according to which subjects could
classify isolated sounds which occurred at random points in time. They either had to ignore the sound (distractor case), or give a
response (target case). They further created acoustic “chimeras”
to investigate which acoustic factor determine source
identification. Therefore they created sounds which had the
spectral envelope of a given sound and the temporal evolution of
another. However, speech sounds, despite sharing many acousticproperties with the “chimeras” could be processed much faster
and more accurately. This points to the role of long-term memory
representations in timbre perception.
Timbre and composition
In the 20th century, many composers have started to experiment with the musical potential of timbre. One of the first was Arnold
Schoenberg, who gave in 1911, on the very last page of his
Harmonielehre, a remarkable quote on the potential role of timbre in composition. Klangfarbenmelodien (tone color melodies / timbremelodies) are musical progressions in which the elements
differentiate from each other by timbres instead of pitches.
In 1909, he had already written music which explores this idea. In
particular in his Farben, op. 16(3), of the 5 pieces for orchestra
makes use of this technique.
Schoenberg’s pupil Arnold Webern played around with
Klangfarbenmelodien as well. One great example of this in his
work is the re-orchestration of Bach’s Ricercata. He introduces a lot of timbral variety in the musical texture! While the traditional
counter-punctual progressions are still important, what gets into
the musical foreground is the diversity and perceptual quality of a texture of ever changing timbres.
Of course, timbre plays a major role not only in contemporary
classical music, but also in “popular” music stemming from the
Rock and Jazz traditions. Here, the electronically manipulated
sounds evoke interest not primarily due to their melodic or
harmonic function, but often more importantly due to their
intriguing timbre.
The perception and memorization of timbre sequences
Schoenberg’s pupil Arnold Webern played around with
Klangfarbenmelodien as well. One great example of this in his
work is the re-orchestration of Bach’s Ricercata. He introduces a lot of timbral variety in the musical texture! While the traditional
counter-punctual progressions are still important, what gets into
the musical foreground is the diversity and perceptual quality of a texture of ever changing timbres.
Of course, timbre plays a major role not only in contemporary
classical music, but also in “popular” music stemming from the
Rock and Jazz traditions. Here, the electronically manipulated
sounds evoke interest not primarily due to their melodic or
harmonic function, but often more importantly due to their
intriguing timbre. Much of the music by Radiohead is a good
example of that.
How exactly timbre is used in such musical situations is not known
and still remains an open and fascinating field of research.
Timbre study I was in was discussed.
Difference in memory capacity@ no difference between musicians non mud for timbre. Equally hard.
