Smoothness Under Parameter Changes: Derivatives and Total Variation

Smoothness Under Parameter Changes: Derivatives and Total Variation

SMOOTHNESS UNDER PARAMETER CHANGES: DERIVATIVES AND TOTAL VARIATION Risto Holopainen ABSTRACT far from smooth responses to changes in physical control variables (e.g. overblowing in wind instruments). Apart from the sounds they make, synthesis models are The smoothness of transitions has been proposed as a cri- distinguished by how the sound is controlled by synthesis terion for evaluating sound morphings [4]. As the mor- parameters. Smoothness under parameter changes is often phing parameter is varied between its extremes, one would a desirable aspect of a synthesis model. The concept of expect the perceived sound to pass through all intermediate smoothness can be made more accurate by regarding the stages as well. However, because of categorical perception synthesis model as a function that maps points in parameter some transitions may not be experienced as gradual. It may space to points in a perceptual feature space. We introduce be impossible to create a convincing morph between, say, new conceptual tools for analyzing the smoothness related a banjo tone and a sustained trombone tone. to the derivative and total variation of a function and apply Quantitative descriptions of the smoothness of a synthesis them to FM synthesis and an ordinary differential equation. parameter should use a measure of the amount of change in The proposed methods can be used to find well behaved the sound, which can be regarded as a distance in a percep- regions in parameter space. tual space. Similarity ratings of pairs of tones have been used in research on timbre perception, where multidimen- 1. INTRODUCTION sional scaling is then used to find a small number of di- mensions that account for the perceived distances between Some synthesis parameters are like switches that can as- stimuli [5]. In several studies, two to four timbral dimen- sume only a discrete set of values, other parameters are like sions have been found and related to various acoustic cor- knobs that can be seamlessly adjusted within some range. relates, often including the attack time, spectral centroid, Only the latter kind of parameter will be discussed here. spectral flux and spectral irregularity [6]. The importance Usually, a small change in some parameter would be ex- of spectrotemporal patterns was stressed in a more recent pected to yield a small change in the sound. As far as this study [7] where five perceptual dimensions were found. is the case, the synthesis model may be said to have well Most timbre studies have focused on pitched, harmonic behaved parameters. sounds, in effect neglecting a large part of the possible A set of criteria for the evaluation of synthesis models range of sounds that can be synthesized. At the other ex- were suggested by Jaffe [1]. Three of the criteria seem rel- treme, the problem of similarity between pieces of music evant in this context: 1) How intuitive are the parameters? has been addressed in music information retrieval [8]. The 2) How perceptible are parameter changes? 3) How well difficulty in comparing two pieces of music is that they behaved are the parameters? The vague notion of smooth- may differ in so many ways, including tempo, instrumen- ness under parameter changes (which is not the name of tation, melodic features and so on. Most synthesis models one of Jaffe’s criteria) can be made more precise by the of interest to musicians are also able to vary along several approach taken in this paper. dimensions of sound, e.g., pitch, loudness, modulation rate From a user’s perspective, the mapping from controllers and many timbral aspects. A thorough study of the per- to synthesis parameters is important [2]. In synthesis mod- ceived changes of sound would include listening tests for els with reasonably well behaved parameters, there are good each synthesis model under investigation. A more tractable prospects of designing mappings that turn the synthesis solution is to use signal descriptors as a proxy for such model and its user interface into a versatile instrument. tests. However, a synthesis model does not necessarily have to There are numerous signal descriptors to choose from [9], have well behaved parameters to be musically useful. De- but the descriptors should respond to parameter changes in spite the counter-intuitive parameter dependencies in com- a given synthesis model. For example, in a study of the plicated nonlinear feedback systems, some musicians are timbre perception of a physical model of the clarinet, the using them [3]. Likewise, acoustic instruments may have attack time, spectral centroid and the ratio of odd to even harmonics were found to be the salient parameters [10]. Copyright: c 2013 Risto Holopainen et al. This is Since a synthesis model may be well behaved with re- an open-access article distributed under the terms of the spect to certain perceptual dimensions but not to others, Creative Commons Attribution 3.0 Unported License, which permits unre- the smoothness may be assessed individually for each of a stricted use, distribution, and reproduction in any medium, provided the original set of complementary signal descriptors. author and source are credited. A synthesis model will be thought of as a function that maps a set of parameter values to a one-sided sequence of space, and another distance metric is needed for points in 0 real numbers, representing the audio samples. It will be the space of sample sequences. Let dp(c; c ) be a metric 0 assumed that all synthesis parameters are set at the begin- in parameter space, and let ds(X(c);X(c )) be a metric in ning of a note event and remain fixed during the note. Dy- the sequence space. The derivative can then be defined as namically varying parameters can be modelled by an LFO the limit or envelope generator, but for simplicity we will consider d (X(c);X(c + δ)) only synthesis parameters that remain constant over time. lim s (1) The effects of parameter changes may be studied either kδk!0 dp(c; c + δ) locally near a specific point in parameter space, or glob- where δ 2 Rp is some small displacement in parameter ally as a parameter varies throughout some range. The lo- space. The limit, if it exists, is the derivative evaluated at cal perspective leads to a notion of the derivative of a syn- the point c. thesis model, which is developed in section2. Parameter In general, synthesis parameters do not make up a uni- changes over a range of values are better described by the form space. Different parameters play different roles; they total variation, which is introduced in section3. Then, sec- affect the sound subtly or dramatically and may interact so tions4 and5 are devoted to case studies of the smoothness that the effect of one parameter depends on the settings of of FM synthesis and the Rossler¨ attractor. Some applica- other parameters. This makes it hard to suggest a general tions and limitations of the methods are discussed in the distance metric that would be suitable for any synthesis conclusion. model. Our solution will be to consider the effects of vary- ing a single synthesis parameter cj at a time, so the distance 0 0 2. SMOOTHNESS BY DERIVATIVE dp(c; c ) in (1) reduces to cj − cj . Furthermore, consider a scalar valued signal descriptor φ(i)(c) ≡ φ(i)(X(c)) In order to formalize the notion of smoothness, we will for- which itself is a signal that depends on the sample se- mulate a synthesis model explicitly as a function and de- quence and the parameter value. Thus, we arrive at a kind scribe what it means for that function to be smooth. First, of partial derivative evaluated with respect to the parameter we define a suitable version of the derivative. Then, in Sec- (i) cj using a signal descriptor φ , tions 2.2 and 2.3, the practicalities of an implementation are discussed. @φ(i) ◦ G(c) d (φ(i)(c); φ(i)(c + he )) = lim s j (2) 2.1 Definition of the derivative @cj h!0 h p Consider a synthesis model as a function G : R ! RN where ej is the jth unit vector in the parameter space. that maps parameters c 2 Rp to a one-sided sequence of Clearly the magnitude of this derivative depends on the samples xn; n = 0; 1; 2;:::, where the sample sequence specifics of the signal descriptors used and which synthesis will be notated X(c) to indicate its dependence on the pa- parameters are considered. In a finite dimensional space, rameters. Then the question of smoothness under param- all partial derivatives should exist and be continuous for eter changes is related to the degree of change in the se- the derivative to exist. Such a strict concept of derivative quence X(c) as the point c in parameter space varies. In does not make sense in the present context where any num- practice, the distance in the output of the synthesis model ber of different signal descriptors can be employed, so only will be measured through a signal descriptor rather than the partial derivatives (2) will be considered. from the raw output signal. If a distance were to be cal- Before discussing the implementation, let us recall some culated from the signals themselves, two periodic signals intuitive conceptions of the derivative. As William Thurston with identical amplitude and frequency but different phase has pointed out [11], mathematicians understand the deriva- might end up being widely separated according to the met- tive in multiple ways, including the following. ric, despite sounding indistinguishable to the human ear. • The derivative is the slope of a line tangent to the Signal descriptors that are clearly affected by the synthesis graph, if it has a tangent.

View Full Text

Details

  • File Type
    pdf
  • Upload Time
    -
  • Content Languages
    English
  • Upload User
    Anonymous/Not logged-in
  • File Pages
    8 Page
  • File Size
    -

Download

Channel Download Status
Express Download Enable

Copyright

We respect the copyrights and intellectual property rights of all users. All uploaded documents are either original works of the uploader or authorized works of the rightful owners.

  • Not to be reproduced or distributed without explicit permission.
  • Not used for commercial purposes outside of approved use cases.
  • Not used to infringe on the rights of the original creators.
  • If you believe any content infringes your copyright, please contact us immediately.

Support

For help with questions, suggestions, or problems, please contact us