basics.lib
A library of basic elements. Its official prefix is ba
.
References
Conversion Tools
(ba.)samp2sec
Converts a number of samples to a duration in seconds at the current sampling rate (see ma.SR
).
samp2sec
is a standard Faust function.
Usage
samp2sec(n) : _
Where:
n
: number of samples
(ba.)sec2samp
Converts a duration in seconds to a number of samples at the current sampling rate (see ma.SR
).
samp2sec
is a standard Faust function.
Usage
sec2samp(d) : _
Where:
d
: duration in seconds
(ba.)db2linear
dBtolinear value converter. It can be used to convert an amplitude in dB to a linear gain ]0N].
db2linear
is a standard Faust function.
Usage
db2linear(l) : _
Where:
l
: amplitude in dB
(ba.)linear2db
lineatodB value converter. It can be used to convert a linear gain ]0N] to an amplitude in dB.
linear2db
is a standard Faust function.
Usage
linear2db(g) : _
Where:
g
: a linear gain
(ba.)lin2LogGain
Converts a linear gain (01) to a log gain (01).
Usage
lin2LogGain(n) : _
Where:
n
: the linear gain
(ba.)log2LinGain
Converts a log gain (01) to a linear gain (01).
Usage
log2LinGain(n) : _
Where:
n
: the log gain
(ba.)tau2pole
Returns a real pole giving exponential decay.
Note that t60 (time to decay 60 dB) is ~6.91 time constants.
tau2pole
is a standard Faust function.
Usage
_ : smooth(tau2pole(tau)) : _
Where:
tau
: timeconstant in seconds
(ba.)pole2tau
Returns the timeconstant, in seconds, corresponding to the given real,
positive pole in (01).
pole2tau
is a standard Faust function.
Usage
pole2tau(pole) : _
Where:
pole
: the pole
(ba.)midikey2hz
Converts a MIDI key number to a frequency in Hz (MIDI key 69 = A440).
midikey2hz
is a standard Faust function.
Usage
midikey2hz(mk) : _
Where:
mk
: the MIDI key number
(ba.)hz2midikey
Converts a frequency in Hz to a MIDI key number (MIDI key 69 = A440).
hz2midikey
is a standard Faust function.
Usage
hz2midikey(freq) : _
Where:
freq
: frequency in Hz
(ba.)semi2ratio
Converts semitones in a frequency multiplicative ratio.
semi2ratio
is a standard Faust function.
Usage
semi2ratio(semi) : _
Where:
semi
: number of semitone
(ba.)ratio2semi
Converts a frequency multiplicative ratio in semitones.
ratio2semi
is a standard Faust function.
Usage
ratio2semi(ratio) : _
Where:
ratio
: frequency multiplicative ratio
(ba.)cent2ratio
Converts cents in a frequency multiplicative ratio.
Usage
cent2ratio(cent) : _
Where:
cent
: number of cents
(ba.)ratio2cent
Converts a frequency multiplicative ratio in cents.
Usage
ratio2cent(ratio) : _
Where:
ratio
: frequency multiplicative ratio
(ba.)pianokey2hz
Converts a piano key number to a frequency in Hz (piano key 49 = A440).
Usage
pianokey2hz(pk) : _
Where:
pk
: the piano key number
(ba.)hz2pianokey
Converts a frequency in Hz to a piano key number (piano key 49 = A440).
Usage
hz2pianokey(freq) : _
Where:
freq
: frequency in Hz
Counters and Time/Tempo Tools
(ba.)counter
Starts counting 0, 1, 2, 3..., and raise the current integer value at each upfront of the trigger.
Usage
counter(trig) : _
Where:
trig
: the trigger signal, each upfront will move the counter to the next integer
(ba.)countdown
Starts counting down from n included to 0. While trig is 1 the output is n.
The countdown starts with the transition of trig from 1 to 0. At the end
of the countdown the output value will remain at 0 until the next trig.
countdown
is a standard Faust function.
Usage
countdown(n,trig) : _
Where:
n
: the starting point of the countdowntrig
: the trigger signal (1: start atn
; 0: decrease until 0)
(ba.)countup
Starts counting up from 0 to n included. While trig is 1 the output is 0.
The countup starts with the transition of trig from 1 to 0. At the end
of the countup the output value will remain at n until the next trig.
countup
is a standard Faust function.
Usage
countup(n,trig) : _
Where:
n
: the maximum count valuetrig
: the trigger signal (1: start at 0; 0: increase untiln
)
(ba.)sweep
Counts from 0 to period1
repeatedly, generating a
sawtooth waveform, like os.lf_rawsaw
,
starting at 1 when run
transitions from 0 to 1.
Outputs zero while run
is 0.
Usage
sweep(period,run) : _
(ba.)time
A simple timer that counts every samples from the beginning of the process.
time
is a standard Faust function.
Usage
time : _
(ba.)ramp
A linear ramp with a slope of '(+/)1/n' samples to reach the next target value.
Usage
_ : ramp(n) : _
Where:
n
: number of samples to increment/decrement the value by one
(ba.)line
A ramp interpolator that generates a linear transition to reach a target value:
 the interpolation process restarts each time a new and distinct input value is received
 it utilizes 'n' samples to achieve the transition to the target value
 after reaching the target value, the output value is maintained.
Usage
_ : line(n) : _
Where:
n
: number of samples to reach the new target received at its input
(ba.)tempo
Converts a tempo in BPM into a number of samples.
Usage
tempo(t) : _
Where:
t
: tempo in BPM
(ba.)period
Basic sawtooth wave of period p
.
Usage
period(p) : _
Where:
p
: period as a number of samples
(ba.)pulse
Pulses (like 10000) generated at period p
.
Usage
pulse(p) : _
Where:
p
: period as a number of samples
(ba.)pulsen
Pulses (like 11110000) of length n
generated at period p
.
Usage
pulsen(n,p) : _
Where:
n
: pulse length as a number of samplesp
: period as a number of samples
(ba.)cycle
Split nonzero input values into n
cycles.
Usage
_ : cycle(n) : si.bus(n)
Where:
n
: the number of cycles/output signals
(ba.)beat
Pulses at tempo t
.
beat
is a standard Faust function.
Usage
beat(t) : _
Where:
t
: tempo in BPM
(ba.)pulse_countup
Starts counting up pulses. While trig is 1 the output is counting up, while trig is 0 the counter is reset to 0.
Usage
_ : pulse_countup(trig) : _
Where:
trig
: the trigger signal (1: start at next pulse; 0: reset to 0)
(ba.)pulse_countdown
Starts counting down pulses. While trig is 1 the output is counting down, while trig is 0 the counter is reset to 0.
Usage
_ : pulse_countdown(trig) : _
Where:
trig
: the trigger signal (1: start at next pulse; 0: reset to 0)
(ba.)pulse_countup_loop
Starts counting up pulses from 0 to n included. While trig is 1 the output is counting up, while trig is 0 the counter is reset to 0. At the end of the countup (n) the output value will be reset to 0.
Usage
_ : pulse_countup_loop(n,trig) : _
Where:
n
: the highest number of the countup (included) before reset to 0trig
: the trigger signal (1: start at next pulse; 0: reset to 0)
(ba.)pulse_countdown_loop
Starts counting down pulses from 0 to n included. While trig is 1 the output is counting down, while trig is 0 the counter is reset to 0. At the end of the countdown (n) the output value will be reset to 0.
Usage
_ : pulse_countdown_loop(n,trig) : _
Where:
n
: the highest number of the countup (included) before reset to 0trig
: the trigger signal (1: start at next pulse; 0: reset to 0)
(ba.)resetCtr
Function that lets through the mth impulse out of
each consecutive group of n
impulses.
Usage
_ : resetCtr(n,m) : _
Where:
n
: the total number of impulses being splitm
: index of impulse to allow to be output
Array Processing/Pattern Matching
(ba.)count
Count the number of elements of list l.
count
is a standard Faust function.
Usage
count(l)
count((10,20,30,40)) > 4
Where:
l
: list of elements
(ba.)take
Take an element from a list.
take
is a standard Faust function.
Usage
take(P,l)
take(3,(10,20,30,40)) > 30
Where:
P
: position (int, known at compile time, P > 0)l
: list of elements
(ba.)subseq
Extract a part of a list.
Usage
subseq(l, P, N)
subseq((10,20,30,40,50,60), 1, 3) > (20,30,40)
subseq((10,20,30,40,50,60), 4, 1) > 50
Where:
l
: listP
: start point (int, known at compile time, 0: begin of list)N
: number of elements (int, known at compile time)
Note:
Faust doesn't have proper lists. Lists are simulated with parallel compositions and there is no empty list.
Function tabulation
The purpose of function tabulation is to speed up the computation of heavy functions over an interval, so that the computation at runtime can be faster than directly using the function. Two techniques are implemented:

tabulate
computes the function in a table and read the points using interpolation.tabulateNd
is the N dimensions version oftabulate

tabulate_chebychev
uses Chebyshev polynomial approximation
Comparison program example
process = line(50000, r0, r1) <: FXtb,FXch : par(i, 2, maxerr)
with {
C = 0;
FX = sin;
NX = 50;
CD = 3;
r0 = 0;
r1 = ma.PI;
tb(x) = ba.tabulate(C, FX, NX*(CD+1), r0, r1, x).cub;
ch(x) = ba.tabulate_chebychev(C, FX, NX, CD, r0, r1, x);
maxerr = abs : max ~ _;
line(n, x0, x1) = x0 + (ba.time%n)/n * (x1x0);
};
(ba.)tabulate
Tabulate a 1D function over the range [r0, r1] for access via nearestvalue, linear, cubic interpolation. In other words, the uniformly tabulated function can be evaluated using interpolation of order 0 (none), 1 (linear), or 3 (cubic).
Usage
tabulate(C, FX, S, r0, r1, x).(vallincub) : _
C
: whether to dynamically force thex
value to the range [r0, r1]: 1 forces the check, 0 deactivates it (constant numerical expression)FX
: unary function Y=F(X) with one output (scalar function of one variable)S
: size of the table in samples (constant numerical expression)r0
: minimum value of argument xr1
: maximum value of argument x
tabulate(C, FX, S, r0, r1, x).val uses the value in the table closest to x
tabulate(C, FX, S, r0, r1, x).lin evaluates at x using linear interpolation between the closest stored values
tabulate(C, FX, S, r0, r1, x).cub evaluates at x using cubic interpolation between the closest stored values
Example test program
midikey2hz(mk) = ba.tabulate(1, ba.midikey2hz, 512, 0, 127, mk).lin;
process = midikey2hz(ba.time), ba.midikey2hz(ba.time);
(ba.)tabulate_chebychev
Tabulate a 1D function over the range [r0, r1] for access via Chebyshev polynomial approximation.
In contrast to (ba.)tabulate
, which interpolates only between tabulated samples, (ba.)tabulate_chebychev
stores coefficients of Chebyshev polynomials that are evaluated to provide better approximations in many cases.
Two new arguments controlling this are NX, the number of segments into which [r0, r1] is divided, and CD,
the maximum Chebyshev polynomial degree to use for each segment. A rdtable
of size NX*(CD+1) is internally used.
Note that processing r1
the last point in the interval is not safe. So either be sure the input stays in [r0, r1[
or use C = 1
.
Usage
_ : tabulate_chebychev(C, FX, NX, CD, r0, r1) : _
C
: whether to dynamically force the value to the range [r0, r1]: 1 forces the check, 0 deactivates it (constant numerical expression)FX
: unary function Y=F(X) with one output (scalar function of one variable)NX
: number of segments for uniformly partitioning [r0, r1] (constant numerical expression)CD
: maximum polynomial degree for each Chebyshev polynomial (constant numerical expression)r0
: minimum value of argument xr1
: maximum value of argument x
Example test program
midikey2hz_chebychev(mk) = ba.tabulate_chebychev(1, ba.midikey2hz, 100, 4, 0, 127, mk);
process = midikey2hz_chebychev(ba.time), ba.midikey2hz(ba.time);
(ba.)tabulateNd
Tabulate an nD function for access via nearestvalue or linear or cubic interpolation. In other words, the tabulated function can be evaluated using interpolation of order 0 (none), 1 (linear), or 3 (cubic).
The table size and parameter range of each dimension can and must be separately specified. You can use it anywhere you have an expensive function with multiple parameters with known ranges. You could use it to build a wavetable synth, for example.
The number of dimensions is deduced from the number of parameters you give, see below.
Note that processing the last point in each interval is not safe. So either be sure the inputs stay in their respective ranges, or use C = 1
. Similarly for the first point when doing cubic interpolation.
Usage
tabulateNd(C, function, (parameters) ).(vallincub) : _
C
: whether to dynamically force the parameter values for each dimension to the ranges specified in parameters: 1 forces the check, 0 deactivates it (constant numerical expression)function
: the function we want to tabulate. Can have any number of inputs, but needs to have just one output.(parameters)
: sizes, ranges and read values. Note: these need to be in brackets, to make them one entity.
If N is the number of dimensions, we need:
 N times
S
: number of values to store for this dimension (constant numerical expression)  N times
r0
: minimum value of this dimension  N times
r1
: maximum value of this dimension  N times
x
: read value of this dimension
By providing these parameters, you indirectly specify the number of dimensions; it's the number of parameters divided by 4.
The user facing functions are:
tabulateNd(C, function, S, parameters).val
 Uses the value in the table closest to x.
tabulateNd(C, function, S, parameters).lin
 Evaluates at x using linear interpolation between the closest stored values.
tabulateNd(C, function, S, parameters).cub
 Evaluates at x using cubic interpolation between the closest stored values.
Example test program
powSin(x,y) = sin(pow(x,y)); // The function we want to tabulate
powSinTable(x,y) = ba.tabulateNd(1, powSin, (sizeX,sizeY, rx0,ry0, rx1,ry1, x,y) ).lin;
sizeX = 512; // table size of the first parameter
sizeY = 512; // table size of the second parameter
rx0 = 2; // start of the range of the first parameter
ry0 = 2; // start of the range of the second parameter
rx1 = 10; // end of the range of the first parameter
ry1 = 10; // end of the range of the second parameter
x = hslider("x", rx0, rx0, rx1, 0.001):si.smoo;
y = hslider("y", ry0, ry0, ry1, 0.001):si.smoo;
process = powSinTable(x,y), powSin(x,y);
Working principle
The .val
function just outputs the closest stored value.
The .lin
and .cub
functions interpolate in N dimensions.
Multi dimensional interpolation
To understand what it means to interpolate in N dimensions, here's a quick reminder on the general principle of 2D linear interpolation:
 We have a grid of values, and we want to find the value at a point (x, y) within this grid.
 We first find the four closest points (A, B, C, D) in the grid surrounding the point (x, y).
Then, we perform linear interpolation in the xdirection between points A and B, and between points C and D. This gives us two new points E and F. Finally, we perform linear interpolation in the ydirection between points E and F to get our value.
To implement this in Faust, we need N sequential groups of interpolators, where N is the number of dimensions.
Each group feeds into the next, with the last "group" being a single interpolator, and the group before it containing one interpolator for each input of the group it's feeding.
Some examples:
 Our 2D linear example has two interpolators feeding into one.
 A 3D linear interpolator has four interpolators feeding into two, feeding into one.
 A 2D cubic interpolater has four interpolators feeding into one.
 A 3D cubic interpolator has sixteen interpolators feeding into four, feeding into one.
To understand which values we need to look up, let's consider the 2D linear example again. The four values going into the first group represent the four closest points (A, B, C, D) mentioned above.
1) The first interpolator gets:
 The closest value that is stored (A)
 The next value in the x dimension, keeping y fixed (B)
2) The second interpolator gets:
 One step over in the y dimension, keeping x fixed (C)
 One step over in both the x dimension and the y dimension (D)
The outputs of these two interpolators are points E and F. In other words: the interpolated x values and, respectively, the following y values:
 The closest stored value of the y dimension
 One step forward in the y dimension
The last interpolator takes these two values and interpolates them in the y dimension.
To generalize for N dimensions and linear interpolation:
 The first group has 2^(n1) parallel interpolators interpolating in the first dimension.
 The second group has 2^(n2) parallel interpolators interpolating in the second dimension.
 The process continues until the nth group, which has a single interpolator interpolating in the nth dimension.
The same principle applies to the cubic interpolation in nD. The only difference is that there would be 4^(n1) parallel interpolators in the first group, compared to 2^(n1) for linear interpolation.
This is what the mixers
function does.
Besides the values, each interpolator also needs to know the weight of each value in it's output.
Let's call this d
, like in ba.interpolate
. It is the same for each group of interpolators, since it correlates to a dimension.
It's value is calculated the similarly to ba.interpolate
:
 First we prepare a "float table readindex" for that dimension (
id
inba.tabulate
)  If the table only had that dimension and it could read a float index, what would it be.
 Then we
int
the float index to get the value we have stored that is closest to, but lower than the input value; the actual index for that dimension. Ourd
is the difference between the float index and the actual index.
The ids
function calculates the id
for each dimension and inside the mixer
function they get turned into d
s.
Storage method
The elephant in the room is: how do we get these indexes? For that we need to know how the values are stored. We use one big table to store everything.
To understand the concept, let's look at the 2D example again, and then we'll extend it to 3d and the general nD case.
Let's say we have a 2D table with dimensions A and B where: A has 3 values between 0 and 5 and B has 4 values between 0 and 1. The 1D array representation of this 2D table will have a size of 3 * 4 = 12.
The values are stored in the following way:
 First 3 values: A is 0, then 3, then 5 while B is at 0.
 Next 3 values: A changes from 0 to 5 while B is at 1/3.
 Next 3 values: A changes from 0 to 5 while B is at 2/3.
 Last 3 values: A changes from 0 to 5 while B is at 1.
For the 3D example, let's extend the 2D example with an additional dimension C having 2 values between 0 and 2. The total size will be 3 * 4 * 2 = 24.
The values are stored like so:
 First 3 values: A changes from 0 to 5, B is at 0, and C is at 0.
 Next 3 values: A changes from 0 to 5, B is at 1/3, and C is at 0.
 Next 3 values: A changes from 0 to 5, B is at 2/3, and C is at 0.
 Next 3 values: A changes from 0 to 5, B is at 1, and C is at 0.
The last 12 values are the same as the first 12, but with C at 2.
For the general ndimensional case, we iterate through all dimensions, changing the values of the innermost dimension first, then moving towards the outer dimensions.
Read indexes
To get the float read index (id
) corresponding to a particular dimension, we scale the function input value to be between 0 and 1, and multiply it by the size of that dimension minus one.
To understand how we get the readIndex
for .val
, let's work trough how we'd do it in our 2D linear example.
For simplicity's sake, the ranges of the inputs to our function
are both 0 to 1.
Say we wanted to read the value closest to x=0.5
and y=0
, so the id
of x
is 1
(the second value) and the id
of y
is 0 (first value). In this case, the read index is just the id
of x
, rounded to the nearest integer, just like in ba.tabulate
.
If we want to read the value belonging to x=0.5
and y=2/3
, things get more complicated. The id
for y
is now 2
, the third value. For each step in the y
direction, we need to increase the index by 3
, the number of values that are stored for x
. So the influence of the y
is: the size of x
times the rounded id
of y
. The final read index is the rounded id
of x
plus the influence of y
.
For the general nD case, we need to do the same operation N times, each feeding into the next. This operation is the riN
function. We take four parameters: the size of the dimension before it prevSize
, the index of the previous dimension prevIX
, the current size sizeX
and the current id idX
. riN
has 2 outputs, the size, for feeding into the next dimension's prevSize
, and the read index feeding into the next dimension's prevIX
.
The size is the sizeX
times prevSize
. The read index is the rounded idX
times prevSize
added to the prevIX
. Our final readIndex
is the read index output of the last dimension.
To get the read values for the interpolators need a pattern of offsets in each dimension, since we are looking for the read indexes surrounding the point of interest. These offsets are best explained by looking at the code of tabulate2d
, the hardcoded 2D version:
tabulate2d(C,function, sizeX,sizeY, rx0,ry0, rx1,ry1, x,y) =
environment {
size = sizeX*sizeY;
// Maximum X index to access
midX = sizeX1;
// Maximum Y index to access
midY = sizeY1;
// Maximum total index to access
mid = size1;
// Create the table
wf = function(wfX,wfY);
// Prepare the 'float' table read index for X
idX = (xrx0)/(rx1rx0)*midX;
// Prepare the 'float' table read index for Y
idY = ((yry0)/(ry1ry0))*midY;
// table creation X:
wfX =
rx0+float(ba.time%sizeX)*(rx1rx0)
/float(midX);
// table creation Y:
wfY =
ry0+
((float(ba.time(ba.time%sizeX))
/float(sizeX))
*(ry1ry0))
/float(midY);
// Limit the table read index in [0, mid] if C = 1
rid(x,mid, 0) = x;
rid(x,mid, 1) = max(0, min(x, mid));
// Tabulate a binary 'FX' function on a range [rx0, rx1] [ry0, ry1]
val(x,y) =
rdtable(size, wf, readIndex);
readIndex =
rid(
rid(int(idX+0.5),midX, C)
+yOffset
, mid, C);
yOffset = sizeX*rid(int(idY),midY,C);
// Tabulate a binary 'FX' function over the range [rx0, rx1] [ry0, ry1] with linear interpolation
lin =
it.interpolate_linear(
dy
, it.interpolate_linear(dx,v0,v1)
, it.interpolate_linear(dx,v2,v3))
with {
i0 = rid(int(idX), midX, C)+yOffset;
i1 = i0+1;
i2 = i0+sizeX;
i3 = i1+sizeX;
dx = idXint(idX);
dy = idYint(idY);
v0 = rdtable(size, wf, rid(i0, mid, C));
v1 = rdtable(size, wf, rid(i1, mid, C));
v2 = rdtable(size, wf, rid(i2, mid, C));
v3 = rdtable(size, wf, rid(i3, mid, C));
};
// Tabulate a binary 'FX' function over the range [rx0, rx1] [ry0, ry1] with cubic interpolation
cub =
it.interpolate_cubic(
dy
, it.interpolate_cubic(dx,v0,v1,v2,v3)
, it.interpolate_cubic(dx,v4,v5,v6,v7)
, it.interpolate_cubic(dx,v8,v9,v10,v11)
, it.interpolate_cubic(dx,v12,v13,v14,v15)
)
with {
i0 = i4sizeX;
i1 = i5sizeX;
i2 = i6sizeX;
i3 = i7sizeX;
i4 = i51;
i5 = rid(int(idX), midX, C)+yOffset;
i6 = i5+1;
i7 = i6+1;
i8 = i4+sizeX;
i9 = i5+sizeX;
i10 = i6+sizeX;
i11 = i7+sizeX;
i12 = i4+(2*sizeX);
i13 = i5+(2*sizeX);
i14 = i6+(2*sizeX);
i15 = i7+(2*sizeX);
dx = idXint(idX);
dy = idYint(idY);
v0 = rdtable(size, wf, rid(i0 , mid, C));
v1 = rdtable(size, wf, rid(i1 , mid, C));
v2 = rdtable(size, wf, rid(i2 , mid, C));
v3 = rdtable(size, wf, rid(i3 , mid, C));
v4 = rdtable(size, wf, rid(i4 , mid, C));
v5 = rdtable(size, wf, rid(i5 , mid, C));
v6 = rdtable(size, wf, rid(i6 , mid, C));
v7 = rdtable(size, wf, rid(i7 , mid, C));
v8 = rdtable(size, wf, rid(i8 , mid, C));
v9 = rdtable(size, wf, rid(i9 , mid, C));
v10 = rdtable(size, wf, rid(i10, mid, C));
v11 = rdtable(size, wf, rid(i11, mid, C));
v12 = rdtable(size, wf, rid(i12, mid, C));
v13 = rdtable(size, wf, rid(i13, mid, C));
v14 = rdtable(size, wf, rid(i14, mid, C));
v15 = rdtable(size, wf, rid(i15, mid, C));
};
};
In the interest of brevity, we'll stop explaining here. If you have any more questions, feel free to open an issue on faustlibraries and tag @magnetophon.
Selectors (Conditions)
(ba.)if
ifthenelse implemented with a select2. WARNING: since select2
is strict (always evaluating both branches),
the resulting if does not have the usual "lazy" semantic of the C if form, and thus cannot be used to
protect against forbidden computations like divisionbyzero for instance.
Usage
if(cond, then, else) : _
Where:
cond
: conditionthen
: signal selected while cond is trueelse
: signal selected while cond is false
(ba.)ifNc
ifthenelseifthen...elsifthenelse implemented on top of ba.if
.
Usage
ifNc((cond1,then1, cond2,then2, ... condN,thenN, else)) : _
or
ifNc(Nc, cond1,then1, cond2,then2, ... condN,thenN, else) : _
or
cond1,then1, cond2,then2, ... condN,thenN, else : ifNc(Nc) : _
Where:
Nc
: number of branches/conditions (constant numerical expression)condX
: conditionthenX
: signal selected if condX is the 1st true conditionelse
: signal selected if all the cond1condN conditions are false
Example test program
process(x,y) = ifNc((x<y,1, x>y,+1, 0));
or
process(x,y) = ifNc(2, x<y,1, x>y,+1, 0);
or
process(x,y) = x<y,1, x>y,+1, 0 : ifNc(2);
outputs 1
if x<y
, +1
if x>y
, 0
otherwise.
(ba.)ifNcNo
ifNcNo(Nc,No)
is similar to ifNc(Nc)
above but then/else branches have No
outputs.
Usage
ifNcNo(Nc,No, cond1,then1, cond2,then2, ... condN,thenN, else) : sig.bus(No)
Where:
Nc
: number of branches/conditions (constant numerical expression)No
: number of outputs (constant numerical expression)condX
: conditionthenX
: list of No signals selected if condX is the 1st true conditionelse
: list of No signals selected if all the cond1condN conditions are false
Example test program
process(x) = ifNcNo(2,3, x<0, 1,1,1, x>0, 1,1,1, 0,0,0);
outputs 1,1,1
if x<0
, 1,1,1
if x>0
, 0,0,0
otherwise.
(ba.)selector
Selects the ith input among n at compile time.
Usage
selector(I,N)
_,_,_,_ : selector(2,4) : _ // selects the 3rd input among 4
Where:
I
: input to select (int, numbered from 0, known at compile time)N
: number of inputs (int, known at compile time, N > I)
There is also cselector for selecting among complex input signals of the form (real,imag).
(ba.)select2stereo
Select between 2 stereo signals.
Usage
_,_,_,_ : select2stereo(bpc) : _,_
Where:
bpc
: the selector switch (0/1)
(ba.)selectn
Selects the ith input among N at run time.
Usage
selectn(N,i)
_,_,_,_ : selectn(4,2) : _ // selects the 3rd input among 4
Where:
N
: number of inputs (int, known at compile time, N > 0)i
: input to select (int, numbered from 0)
Example test program
N = 64;
process = par(n, N, (par(i,N,i) : selectn(N,n)));
(ba.)selectmulti
Selects the ith circuit among N at run time (all should have the same number of inputs and outputs) with a crossfade.
Usage
selectmulti(n,lgen,id)
Where:
n
: crossfade in sampleslgen
: list of circuitsid
: circuit to select (int, numbered from 0)
Example test program
process = selectmulti(ma.SR/10, ((3,9),(2,8),(5,7)), nentry("choice", 0, 0, 2, 1));
process = selectmulti(ma.SR/10, ((_*3,_*9),(_*2,_*8),(_*5,_*7)), nentry("choice", 0, 0, 2, 1));
(ba.)selectoutn
Route input to the output among N at run time.
Usage
_ : selectoutn(N, i) : si.bus(N)
Where:
N
: number of outputs (int, known at compile time, N > 0)i
: output number to route to (int, numbered from 0) (i.e. slider)
Example test program
process = 1 : selectoutn(3, sel) : par(i, 3, vbargraph("v.bargraph %i", 0, 1));
sel = hslider("volume", 0, 0, 2, 1) : int;
Other
(ba.)latch
Latch input on positivegoing transition of trig:"records" the input when trig switches from 0 to 1, outputs a frozen values everytime else.
Usage
_ : latch(trig) : _
Where:
trig
: hold trigger (0 for hold, 1 for bypass)
(ba.)sAndH
Sample And Hold: "records" the input when trig is 1, outputs a frozen value when trig is 0.
sAndH
is a standard Faust function.
Usage
_ : sAndH(trig) : _
Where:
trig
: hold trigger (0 for hold, 1 for bypass)
(ba.)downSample
Down sample a signal. WARNING: this function doesn't change the
rate of a signal, it just holds samples...
downSample
is a standard Faust function.
Usage
_ : downSample(freq) : _
Where:
freq
: new rate in Hz
(ba.)peakhold
Outputs current max value above zero.
Usage
_ : peakhold(mode) : _
Where:
mode
means:
0  Pass through. A single sample 0 trigger will work as a reset.
1  Track and hold max value.
(ba.)peakholder
While peakholder functions are scarcely discussed in the literature (please do send me an email if you know otherwise), common sense tells that the expected behaviour should be as follows: the absolute value of the input signal is compared with the output of the peakholder; if the input is greater or equal to the output, a new peak is detected and sent to the output; otherwise, a timer starts and the current peak is held for N samples; once the timer is out and no new peaks have been detected, the absolute value of the current input becomes the new peak.
Usage
_ : peakholder(holdTime) : _
Where:
holdTime
: hold time in samples
(ba.)impulsify
Turns a signal into an impulse with the value of the current sample
(0.3,0.2,0.1 becomes 0.3,0.0,0.0). This function is typically used with a
button
to turn its output into an impulse. impulsify
is a standard Faust
function.
Usage
button("gate") : impulsify;
(ba.)automat
Record and replay in a loop the successives values of the input signal.
Usage
hslider(...) : automat(t, size, init) : _
Where:
t
: tempo in BPMsize
: number of items in the loopinit
: init value in the loop
(ba.)bpf
bpf is an environment (a group of related definitions) that can be used to create breakpoint functions. It contains three functions:
start(x,y)
to start a breakpoint functionend(x,y)
to end a breakpoint functionpoint(x,y)
to add intermediate points to a breakpoint function
A minimal breakpoint function must contain at least a start and an end point:
f = bpf.start(x0,y0) : bpf.end(x1,y1);
A more involved breakpoint function can contains any number of intermediate points:
f = bpf.start(x0,y0) : bpf.point(x1,y1) : bpf.point(x2,y2) : bpf.end(x3,y3);
In any case the x_{i}
must be in increasing order (for all i
, x_{i} < x_{i+1}
).
For example the following definition:
f = bpf.start(x0,y0) : ... : bpf.point(xi,yi) : ... : bpf.end(xn,yn);
implements a breakpoint function f such that:
f(x) = y_{0}
whenx < x_{0}
f(x) = y_{n}
whenx > x_{n}
f(x) = y_{i} + (y_{i+1}y_{i})*(xx_{i})/(x_{i+1}x_{i})
whenx_{i} <= x
andx < x_{i+1}
bpf
is a standard Faust function.
(ba.)listInterp
Linearly interpolates between the elements of a list.
Usage
index = 1.69; // range is 04
process = listInterp((800,400,350,450,325),index);
Where:
index
: the index (float) to interpolate between the different values. The range ofindex
depends on the size of the list.
(ba.)bypass1
Takes a mono input signal, route it to e
and bypass it if bpc = 1
.
When bypassed, e
is feed with zeros so that its state is cleanup up.
bypass1
is a standard Faust function.
Usage
_ : bypass1(bpc,e) : _
Where:
bpc
: bypass switch (0/1)e
: a mono effect
(ba.)bypass2
Takes a stereo input signal, route it to e
and bypass it if bpc = 1
.
When bypassed, e
is feed with zeros so that its state is cleanup up.
bypass2
is a standard Faust function.
Usage
_,_ : bypass2(bpc,e) : _,_
Where:
bpc
: bypass switch (0/1)e
: a stereo effect
(ba.)bypass1to2
Bypass switch for effect e
having mono input signal and stereo output.
Effect e
is bypassed if bpc = 1
.When bypassed, e
is feed with zeros
so that its state is cleanup up.
bypass1to2
is a standard Faust function.
Usage
_ : bypass1to2(bpc,e) : _,_
Where:
bpc
: bypass switch (0/1)e
: a monotostereo effect
(ba.)bypass_fade
Bypass an arbitrary (N x N) circuit with 'n' samples crossfade.
Inputs and outputs signals are faded out when 'e' is bypassed,
so that 'e' state is cleanup up.
Once bypassed the effect is replaced by par(i,N,_)
.
Bypassed circuits can be chained.
Usage
_ : bypass_fade(n,b,e) : _
or
_,_ : bypass_fade(n,b,e) : _,_
n
: number of samples for the crossfadeb
: bypass switch (0/1)e
: N x N circuit
Example test program
process = bypass_fade(ma.SR/10, checkbox("bypass echo"), echo);
process = bypass_fade(ma.SR/10, checkbox("bypass reverb"), freeverb);
(ba.)toggle
Triggered by the change of 0 to 1, it toggles the output value between 0 and 1.
Usage
_ : toggle : _
Example test program
button("toggle") : toggle : vbargraph("output", 0, 1)
(an.amp_follower(0.1) > 0.01) : toggle : vbargraph("output", 0, 1) // takes audio input
(ba.)on_and_off
The first channel set the output to 1, the second channel to 0.
Usage
_,_ : on_and_off : _
Example test program
button("on"), button("off") : on_and_off : vbargraph("output", 0, 1)
(ba.)bitcrusher
Produce distortion by reduction of the signal resolution.
Usage
_ : bitcrusher(nbits) : _
Where:
nbits
: the number of bits of the wanted resolution
Sliding Reduce
Provides various operations on the last n samples using a high order
slidingReduce(op,n,maxN,disabledVal,x)
foldlike function:
slidingSum(n)
: the sliding sum of the last n input samples, CPUlightslidingSump(n,maxN)
: the sliding sum of the last n input samples, numerically stable "forever"slidingMax(n,maxN)
: the sliding max of the last n input samplesslidingMin(n,maxN)
: the sliding min of the last n input samplesslidingMean(n)
: the sliding mean of the last n input samples, CPUlightslidingMeanp(n,maxN)
: the sliding mean of the last n input samples, numerically stable "forever"slidingRMS(n)
: the sliding RMS of the last n input samples, CPUlightslidingRMSp(n,maxN)
: the sliding RMS of the last n input samples, numerically stable "forever"
Working Principle
If we want the maximum of the last 8 values, we can do that as:
simpleMax(x) =
(
(
max(x@0,x@1),
max(x@2,x@3)
) :max
),
(
(
max(x@4,x@5),
max(x@6,x@7)
) :max
)
:max;
max(x@2,x@3)
is the same as max(x@0,x@1)@2
but the latter reuses a
value we already computed,so is more efficient. Using the same trick for
values 4 trough 7, we can write:
efficientMax(x)=
(
(
max(x@0,x@1),
max(x@0,x@1)@2
) :max
),
(
(
max(x@0,x@1),
max(x@0,x@1)@2
) :max@4
)
:max;
We can rewrite it recursively, so it becomes possible to get the maximum at have any number of values, as long as it's a power of 2.
recursiveMax =
case {
(1,x) => x;
(N,x) => max(recursiveMax(N/2,x), recursiveMax(N/2,x)@(N/2));
};
What if we want to look at a number of values that's not a power of 2?
For each value, we will have to decide whether to use it or not.
If n is bigger than the index of the value, we use it, otherwise we replace
it with (0(ma.MAX)
):
variableMax(n,x) =
max(
max(
(
(x@0 : useVal(0)),
(x@1 : useVal(1))
):max,
(
(x@2 : useVal(2)),
(x@3 : useVal(3))
):max
),
max(
(
(x@4 : useVal(4)),
(x@5 : useVal(5))
):max,
(
(x@6 : useVal(6)),
(x@7 : useVal(7))
):max
)
)
with {
useVal(i) = select2((n>=i) , (0(ma.MAX)),_);
};
Now it becomes impossible to reuse any values. To fix that let's first look
at how we'd implement it using recursiveMax, but with a fixed n that is not
a power of 2. For example, this is how you'd do it with n=3
:
binaryMaxThree(x) =
(
recursiveMax(1,x)@0, // the first x
recursiveMax(2,x)@1 // the second and third x
):max;
n=6
binaryMaxSix(x) =
(
recursiveMax(2,x)@0, // first two
recursiveMax(4,x)@2 // third trough sixth
):max;
Note that recursiveMax(2,x)
is used at a different delay then in
binaryMaxThree
, since it represents 1 and 2, not 2 and 3. Each block is
delayed the combined size of the previous blocks.
n=7
binaryMaxSeven(x) =
(
(
recursiveMax(1,x)@0, // first x
recursiveMax(2,x)@1 // second and third
):max,
(
recursiveMax(4,x)@3 // fourth trough seventh
)
):max;
To make a variable version, we need to know which powers of two are used, and at which delay time.
Then it becomes a matter of:
 lining up all the different block sizes in parallel:
sequentialOperatorParOut()
 delaying each the appropriate amount:
sumOfPrevBlockSizes()
 turning it on or off:
useVal()
 getting the maximum of all of them:
parallelOp()
In Faust, we can only do that for a fixed maximum number of values: maxN
, known at compile time.
(ba.)slidingReduce
Foldlike high order function. Apply a commutative binary operation op
to
the last n
consecutive samples of a signal x
. For example :
slidingReduce(max,128,128,0(ma.MAX))
will compute the maximum of the last
128 samples. The output is updated each sample, unlike reduce, where the
output is constant for the duration of a block.
Usage
_ : slidingReduce(op,n,maxN,disabledVal) : _
Where:
n
: the number of values to processmaxN
: the maximum number of values to process (int, known at compile time, maxN > 0)op
: the operator. Needs to be a commutative one.disabledVal
: the value to use when we want to ignore a value.
In other words, op(x,disabledVal)
should equal to x
. For example,
+(x,0)
equals x
and min(x,ma.MAX)
equals x
. So if we want to
calculate the sum, we need to give 0 as disabledVal
, and if we want the
minimum, we need to give ma.MAX
as disabledVal
.
(ba.)slidingSum
The sliding sum of the last n input samples.
It will eventually run into numerical trouble when there is a persistent dc component.
If that matters in your application, use the more CPUintensive ba.slidingSump
.
Usage
_ : slidingSum(n) : _
Where:
n
: the number of values to process
(ba.)slidingSump
The sliding sum of the last n input samples.
It uses a lot more CPU than ba.slidingSum
, but is numerically stable "forever" in return.
Usage
_ : slidingSump(n,maxN) : _
Where:
n
: the number of values to processmaxN
: the maximum number of values to process (int, known at compile time, maxN > 0)
(ba.)slidingMax
The sliding maximum of the last n input samples.
Usage
_ : slidingMax(n,maxN) : _
Where:
n
: the number of values to processmaxN
: the maximum number of values to process (int, known at compile time, maxN > 0)
(ba.)slidingMin
The sliding minimum of the last n input samples.
Usage
_ : slidingMin(n,maxN) : _
Where:
n
: the number of values to processmaxN
: the maximum number of values to process (int, known at compile time, maxN > 0)
(ba.)slidingMean
The sliding mean of the last n input samples.
It will eventually run into numerical trouble when there is a persistent dc component.
If that matters in your application, use the more CPUintensive ba.slidingMeanp
.
Usage
_ : slidingMean(n) : _
Where:
n
: the number of values to process
(ba.)slidingMeanp
The sliding mean of the last n input samples.
It uses a lot more CPU than ba.slidingMean
, but is numerically stable "forever" in return.
Usage
_ : slidingMeanp(n,maxN) : _
Where:
n
: the number of values to processmaxN
: the maximum number of values to process (int, known at compile time, maxN > 0)
(ba.)slidingRMS
The root mean square of the last n input samples.
It will eventually run into numerical trouble when there is a persistent dc component.
If that matters in your application, use the more CPUintensive ba.slidingRMSp
.
Usage
_ : slidingRMS(n) : _
Where:
n
: the number of values to process
(ba.)slidingRMSp
The root mean square of the last n input samples.
It uses a lot more CPU than ba.slidingRMS
, but is numerically stable "forever" in return.
Usage
_ : slidingRMSp(n,maxN) : _
Where:
n
: the number of values to processmaxN
: the maximum number of values to process (int, known at compile time, maxN > 0)
Parallel Operators
Provides various operations on N parallel inputs using a high order
parallelOp(op,N,x)
function:
parallelMax(N)
: the max of n parallel inputsparallelMin(N)
: the min of n parallel inputsparallelMean(N)
: the mean of n parallel inputsparallelRMS(N)
: the RMS of n parallel inputs
(ba.)parallelOp
Apply a commutative binary operation op
to N parallel inputs.
usage
si.bus(N) : parallelOp(op,N) : _
where:
N
: the number of parallel inputs known at compile timeop
: the operator which needs to be commutative
(ba.)parallelMax
The maximum of N parallel inputs.
Usage
si.bus(N) : parallelMax(N) : _
Where:
N
: the number of parallel inputs known at compile time
(ba.)parallelMin
The minimum of N parallel inputs.
Usage
si.bus(N) : parallelMin(N) : _
Where:
N
: the number of parallel inputs known at compile time
(ba.)parallelMean
The mean of N parallel inputs.
Usage
si.bus(N) : parallelMean(N) : _
Where:
N
: the number of parallel inputs known at compile time
(ba.)parallelRMS
The RMS of N parallel inputs.
Usage
si.bus(N) : parallelRMS(N) : _
Where:
N
: the number of parallel inputs known at compile time