To overcome these limitations we have resorted to other software to generate the Csound scores: AC Toolbox (Berg, 1998), Alcaucil (Keller, 1998). Nevertheless, CMask has proven to be useful for synthesizing several sound classes. The working method that we present in this tutorial covers a few classes of ecologically-modeled sounds, and provides some basic examples of the ecological approach to sound resynthesis. No attempt is made to cover theoretical and conceptual issues which are already discussed in other sources (cf. Keller & Truax, 1998; Keller & Rolfe, 1998; Keller, 1999b).

Throughout this tutorial, we use the word event to define a sound produced by the combination of multiple grains (in the order of hundreds to thousands). The grains are previously sampled and edited sounds which keep their original spectral and temporal qualities (Keller & Truax, 1998). Thus, a single event is generated by this procedure:

• Select and edit a number of sound samples (source files) in a sound editor.
  
  

We have tried to keep some consistency by following Csound syntax conventions in the score formatting process. Resolution is always kept at 5 decimal places for floating point variables, and at integer for integer variables (usually used for ftable numbers). In CMask syntax it is indicated thus:

prec 5    ; 5 decimal places (0.00000). prec 0    ; integer.

Most parameter values are normalized (0 to 1 range) and scaled in the Csound instrument. We keep the sound file¹s original amplitude and select the scaling option in Csound to avoid clipping.

Each CMask parameter stands for a pfield in Csound. p1 to p3 are mandatory to get any output from CMask. To simplify notation, we have not included these parameters in the examples. So the user should add them when trying out the examples. All our Csound instruments read each line score by this convention (p stands for pfield or parameter field):

(mono) p1 instrument number p2 onset time p3 duration p4 amplitude p5 frequency p6 sound file

(stereo) p1 instrument number p2 onset time p3 duration p4 amplitude p5 frequency left channel p6 frequency right channel p7 sound file left channel p8 sound file right channel

ftables provide the fastest way to keep sounds available in RAM. When using stereo files, care should be taken to keep the stereo image by using consistent playback of the two channels (of course, sometimes this is not wanted or needed). Here is a typical ftable statement:

f2 0 65536 -1 ³soundfile.sd2² 0 4 1 ; left channel f3 0 65536 -1 ³soundfile.sd2² 0 4 2 ; right channel

We reserve ftable 1 for the windowing function. While using several dozens (or hundreds) of sound files, we felt the need to automate this process a bit. So we made some Applescripts for BBEdit 4.0 to write the file names and the trapezoidal window function.

There are three types of interpolation available in CMask (we encourage to use always one of them):

ipl 0 ipl 1.5 ipl -1 ipl cos

; linear: a straight line between initial and end value. ; exponential: a convex curve when the sign is positive. ; a concave curve when the sign is negative. ; half-cosine: useful when doing smooth transitions between adjoining events.

There are two possible uses of random functions in CMask: (1) fixed range, (2) time-varying range. In the fixed range case, the low and high boundaries are kept constant throughout the event duration. For the time-varying range a tendency mask is employed.


Example 1

f 0 5 p1 range .1 .3 prec 5

; Fixed range ; random function. ; low boundary = .1, high boundary = .3.

f 0 5 p1 range .1 2 mask [.1 1] [.3 2] prec 5

; Time-varying range ; random function. ; initial low boundary = .1 ; initial high boundary = .3, ; final low boundary = 1 ; initial high boundary = 2

In the fixed range example, the grain onsets will be constantly spread between 10 and 30 ms. In the second example, the density will begin at a 10 to 30 ms. range and will finish at a range of 1 to 2 seconds once the five-second event duration has elapsed.

There are two methods in CMask to establish transitions from random values to deterministic patterns. One is to use a tendency mask to reduce the range until a single value is reached. Another one is to set a grid of values using quant. Then change the degree of randomness around the deterministic grid.


Example 2

f 0 5 p5 range .8 1.2 mask [.8 1] [1.2 1] prec 5

In this example, the playback frequency of the sound file is randomly varied within a range of 20% higher and 20% lower sample rates. This range is shrunk until the original frequency is reached after 5 seconds.


Example 3

f 0 5 p4 range 0 1 quant .2 [1 0] .1 prec 5

; random function, range = 0-1 ; grid parameters: interval = .2 ; strength = 0 to 1 ; offset = .1

In this case, we set a grid for the amplitude values at an interval of .2, that is (0, .2, .4, .6, .8, 1). Given that the strength parameter is set to maximum, 1, the only amplitudes values that we¹ll get are the ones in the grid. Let¹s say that we do not want any silence (amplitude = 0). Then we set the third parameter, offset, to .1. This adds .1 to all values so we will get values from .1 to 1.1, and the grid will become (.1, .3, .5, .7, .9, 1.1). As time passes, the grid strength is reduced until reaching 0. At 5 seconds, the grid disappears and we are left with a random distribution.

Most ecological sound models are constrained to a small variation in their time-spans. Their timbral characteristics are directly linked to their evolution through time. By keeping parameters linked to the time-axis we allow for changes in their time-scale without affecting their mesostructure and without having to remap parameters. In simple words, bubble-gum functions work as a bandoneon. They adapt to the Œlength¹ they are given. These functions are defined by:

When synthesizing complex events, such as breaking glass or multi-bouncing patterns, sometimes we need to combine several sub-events. When a seamless transition is desired half-cosine interpolation can be used to concatenate the grain distribution of two sub-events. Extending the previous example:


Example 5

f 0 5 p4 range 0 1 mask [0 1 ipl cos] [.1 1 ipl cos] prec 5

; first sub-event. ; half-cosine interpolation.

f 5 10 p4 range 0 1 mask [1 0 ipl cos] [1 .1 ipl cos] prec 5

; second sub-event. ; half-cosine interpolation.

By combining two sub-events of 5 seconds, we get a ten-second event with an amplitude profile going from 0-.1 to 1-1 and back to 0-.1. No discontinuity or sudden peak will be produced at 5 seconds.

accum, mask, and quant have been widely used in ecological models. So we provide some CMask examples to show how they can be employed in simple tasks. accum has been used to implement a scraper model. quant allowed us to produce a continuum of voice-like clusters which slowly turn into a rich harmonic sound (cf. Vox Populi in touch¹n¹go).


Example 6

accum accumtest.msk accumtest.orc accumtest1.msk Samples bot.sd2 metal.sd2

quant quantest.msk quantest.orc quantest1.msk quantest2.msk quantest3.msk

mask mask2events.msk mask2events.orc masktest.msk masktest.orc masktest1.msk masktest2.msk masktest3.msk masktest4.msk masktest5.msk

(Reminder: when synthesizing the examples, do not forget to place the two samples provided, in the Csound sample folder. The samples are in Sound Designer II format.)