27/05/2022 document based on version 0.10 of Syntə


Syntə is an audio live coding environment

The name is pronounced ‘sinter‘, which means to create something by fusing many tiny elements together under intense heat. It is also a portmanteau of ‘synth’ and ‘byte’, which is a reference to the stream of bytes that are generated and sent to the soundcard by Syntə

The input syntax is in EBNF: operator [ ” ” operand ] Where an operand can be a name or a number where: name = letter { letter | digit } number = float [ type ] [ ( “/” | “*” ) float [type] A letter is defined as any UTF-8 character excluding + – . 0 1 2 3 4 5 6 7 8 9 A float matches the floating point literal in the Go language specification. A type can be one of the following tokens: “hz”, “s”, “ms”, “bpm”, “!”, or “db” . A list of operators is given below. Lists of operations may be composed into functions with multiple arguments. The function syntax is = function [ ” ” operand [ “,” operand ] [ “,” operand ] ] .

Protect your hearing when listening to a system capable of more than 85dB SPL




Audio synthesis √ Wav playback √ Mouse control √ Telemetry / code display √ Finite recursion with enumeration ◊ Anything can be connected to anything else within a listing √ Feedback permitted (see above) √ Groups of operators can be defined, named and instantiated as functions (extensible) √ Support for pitch control with useful constants √ ◊ Frequency scaling √ Useful predefined functions √ Built-in synchronisation operators √

Intended purpose:

To compose and perform music using algorithms

Author: Dan Arves

Available for workshops, talks and performances: [email protected]

This work and associated code is not currently licensed © 2022

For now this document also serves as an (incomplete) specification of the syntə language and may be viewed as a paper on the topic.

This document has been written to be accessible to the widest audience as a deliberate aim. Some prior knowledge of unix-like sytems and sound synthesis will be useful, and you can build up some knowledge about them alongside this document. Syntə is designed to be both capable as a serious sonic tool and a good entry point for beginners, although inevitably there is a tradeoff and so some patience and learning may be required if you are starting from scratch.

The ◊ symbol indicates a sentence or section that may need updating in future.

How to use syntə


Computer with a soundcard (internal or external) OSS sound driver (FreeBSD) ◊ Go programming language installed ◊ Desire to learn about audio synthesis

Linux and ALSA driver will work with some simple modifications, use the synte.linux.go file if available. It is not known at present what the performance will be on other systems. ◊ The terminal emulator that has been used for development and testing is Alacritty. It works well on cool-retro-term. You may experience flickering in some terminal emulators due to the incomplete UI.

The source code is as yet unlicensed so be wary of distributing or hosting the code, as-is, compiled, modified or otherwise. ◊

Create or navigate to a directory (folder) containing the following files:

an empty directory named 'recordings'   
an empty file named 'displaylisting.json' ◊ (only empty on first use) 
a directory named 'wavs' containing wav files (optional)

Open a terminal and type ‘go run synte.go’ to begin. ◊ Open another terminal and run info.go similarly. This will display useful information and feedback as you input and run code. Open another terminal and run listing.go to view currently running code, this will also show mute status in italics. You may wish to arrange these using a tiling window manager, terminal multiplexer, or equivalent.

You will be prompted to write your first syntə listing, a program that will make sounds. The listing is input one line at a time. You must write the name of an operator or function, usually followed by a space and a number or signal name. The number must conform to Go’s float format or it will be assumed to be a name. The first character of a name cannot be a number, plus, minus or dot, to avoid confusion. Some names have special meaning, such as ones beginning with the @ or ^ characters.

Press enter to complete each line.

Try this example to test everything is working ok:

in 330hz  
mul 0.2  
out dac  

This should output a single sine tone.

To exit from syntə type : exit The : operator is used for commands to syntə

A function is a predefined list of operations that you can use in your code as if they are operators. These are indicated by a different colour once you have typed them in.

The result of each operation is fed to the input of the next. You can call this a necklace, or listing of operations. One link of the necklace looks like this in schematic form:

input (from previous)
[operator] [operand]
output (to next)

A number represents an input value that could be used to set, for example, a frequency. A number can be input as a frequency, time, or bpm or just a plain number. Eg. in 1/3, in 330hz, in 2s, in 500ms, in 120bpm, mul 6db. A name represents a signal, Signals are used to store or share values using the in and out operators. Another way to think of them is as registers, like in a CPU. You can only output to one unique signal, but you can input multiple times from the same signal. Think of it as a confluence of rivers flowing into one another to reach the sea. Many inputs, one output. If you want to send two values to the same signal, either add them or combine them in other ways using operators first. A signal name that isn’t visible if it is within a function in the necklace will still count as a duplicate. For this reason a list of signals is displayed in the terminal running info.go . Most signals within functions are labelled a, b or c so try the next letter of the alphabet instead.

Some operators and functions do not need an operand, simply press enter after the name.

Common operations are +, mul, in, out.

The osc function outputs a ramp wave (increasing series of values up to 1, then restarts.) osc accepts a frequency in hertz (from the preceding operation.)

Most values apart from frequencies are between -1 and 1 for audio and 0 to 1 for control/modulation. This might seem like a limited range, however incredibly small fractions down to a number with over 300 zeros after the decimal place are possible. You won’t need to handle such numbers, but if you want to experiment they can be input using the syntax 1e-3 which would represent one thousandth, or 3 millionths would be 3e-6 etc. You may also input a number as a fraction such as in 1/3.14 etc. Any values greater than 1 or less than -1 will be clipped by the output resulting in distortion. Think of this as slicing off the tops of waveforms that are too loud. Most of the time you won’t need to worry about keeping within range though.

In the syntə code example above, the output of the osc function is shaped by the sine operator. The sine operator does not produce a sound tone by itself.

dac is a special signal name which sends the output to the soundcard of your computer. Typing out dac ends a listing and sends it to the sound engine to perform computations. All listings end in out dac, so it can only be used once. dac stands for digital audio converter, which is the technical name for anything that takes a series of digitally represented numbers and outputs them as audio.

You will probably want to use the function mix which contains out dac and so will also end a necklace. mix will set the level based on the frequency of the most recent osc function to ensure output limiting doesn’t take place.

The sino function combines osc and sine so can be used in their place.

Once you have added one listing in this way, you can add more. Each listing has a number indicated at the beginning of the first line. This is used to reference the listing for operators like del which silences a listing and removes all its code. Typing : erase removes all operations input to a listing so far and starts again from the top.

You don’t need to understand all of this right away. Everyone learns at different rates, and has a different learning style. Some need to experiment, some need a lot of detail, some by example etc. In future you may be able to find a workshop to attend if that is helpful too. If you want you can skip to the examples section below to try out some listings directly.

Great, but what do I actually type in?

To decide what code to write, it’s helpful to think about the functions or shapes that represent the sounds that you want. Or you may want to skip this next section for now and head to the examples in the next section.

Any number that increases and decreases periodically in some way will produce a an audible sound if the frequency is in the range of 20 to 20,000 hertz. (1 hertz is one cycle per second.) In practice most frequencies will be betwen around 200 to 2000 hertz. Anything below that takes quite a lot of power and size for a speaker to produce clearly and anything above that is rarely used directly (as a fundamental or root note) and will usually only appear as overtones of other lower frequencies. A discussion of overtones and harmonics is beyond the scope of this document, doing your own research will be useful for future if they are new to you.

The osc function is actually a group of operators that are added to your listing. They are:

out ^freq  
+ a  
mod 1  
out a  

out ^freq sends the input to a mix function (if you use one) for help in setting a sensible level. + adds the input to a signal called a . mod 1 allows through any number between 0 and 1. Any numbers larger have 1 subtracted until within that range, eg. 2.3 → 0.3 out a sends the result of mod 1 to the signal a, this forms a loop.

So osc is always adding to itself, yet keeping between 0 and 1. In ASCII it looks like: /|/|/| etc, you can see why it is called a ramp wave. In the study of digital signal processing it is known as a phase accumulator and in mathematics it is an overflowing integrator, but you don’t need to know that. Another way of looking at it is that it counts up and then starts again to generate a repeating cycle. It is one of the most useful and basic functions in producing sounds. In fact, unless you do something really fiddly, nearly all of your necklaces will contain osc somewhere. (For functional programming afficionados the register in osc forms a closure which holds the current value of the output. Any register that is read from with in before writing to with out will form a closure, if this is allowed to change while being constrained within a certain range it will form an oscillator.)

Now what happens if you want a decreasing wave form? i.e one that counts down instead of up. What you can do is take an osc and flip it upside down with mul -1. Now it is also negative so you can + 1 to shift it between 0 and 1 again. This is also what the flip function does. This upside down ramp can be called a sawtooth wave (although they’re kind of interchangable terms.) You can also use the saw function which encapsulates osc and flip to save typing. Interestingly you wouldn’t be able to tell an audible difference betweeen a ramp and a sawtooth wave if output directly to your speakers, that is to say they sound the same. However for low frequency modulation, like periodically changing the volume, you would definitely tell the difference. Modulation just means changing one aspect of something else, such as the volume or filtering or pitch.

Signal chains

If you are not familiar with the concepts of synthesis don’t worry, with a little time and patience you can build up confidence and understanding bit by bit.

To understand signal chains, visualise how a traditional electric guitarist makes music. They connect their equipment like so:

Guitar → fx pedal → amplifier

This is the equivalent to a necklace in syntə, for example something like:

frequency → osc → filt → out dac

But in this case syntə is the guitar and the guitarist too. So in code in 330, osc, osc, out dac would be a frequency controlling the frequency of another oscillator. In this case the frequency of the second oscillator will only change from 0 to 1, so we can:

in 330hz, osc, mul f, osc, out dac

where f is now the maximum frequency we span to. You can use a number such as 200 instead, or feed in a value by sending to f from somewhere else. The listing display indicates where a result is passed down the chain by a curly arrow on the left. The operators in, pop, and tap break the chain and start a with value which is either the operand, the top of the stack, or the specified delay tap, respectively.

If we examine the guitarist example more closely , we realise that the fx pedal will likely contain modulators such as LFOs, along with filters, wave shapers etc. So we can extend our analogous model to something like:

in 2.5, osc, inv, out b, tri, out c, in 330hz, osc, mul b, filt c, mix

Notice that the first oscillator is now generating a 2.5 hertz low frequency modulation. It is inverted before being send to b, and then shaped into a triangle wave and send to c. The second osc now operates at a fixed frequency of 330 hertz and is multiplied by b (acting like a VCA to control the volume) and filtered by c before being mixed and send to the output.

Which leads us to


To assist in writing necklaces of operations here are some useful heuristics/suggestions:

  • Low to high. Start from low frequencies (less than 20 hertz) for modulation and combine them and/or outputting to signals with out for later use, before progressing to high (audible) frequencies.

  • name each signal by the next available letter of the alphabet. Added functions will contain signals that are numbered to avoid conflicts, these are displayed in the info.go terminal so you can avoid them. ◊ Another approach, which is good for beginners is semantic naming, which is a clever way of saying to give a name that describes what the value represents – what it means. For example: OscPitch, modulatorA, lfo, env1 etc.

  • if you realise that you need to add in further modulation or adjust values, you can use push to break the necklace and then pop to resume later.

Examples ◊

To help get your creative juices flowing, try typing in these example necklaces of operations:

Surf Add two or three separate listings of this code for a relaxing beach experience.

in 4hz
+ 0.05hz
out b
filt b

The first out register is named b because a is already in use within the osc function

Kick and hihat

in 2hz (or 120bpm)
out b
mul 55hz
mul b
mul 3

The kick will play on every beat. For once per bar of four beats use in 120bpm, / 4 before posc

in 4hz
posc 0.5
+ -1.5
+ 1

in -1

Here the clip operator is used to shape the VCA envelope of the hi-hat and the listings are synchronised together with a phase offset.

Sample and hold melody

in 3.5hz
out b
in 8hz
s/h b
mul 3.5
mul 8
mod 3
mul ln3
mod ln2
base E
mul 440

Here the pitch is calculated exponentially, mixing powers of 3 ≡ MOD 2 to produce a scale. You are not expected to understand this straightaway! The bpm could be interpreted as quarter notes at 120bpm, because 8 / 4 = 2 and 2 x 60 = 120.

Pulse sequencing

in 120bpm
mul 1/4
out tempo
pulse 2/4
out pitch
in tempo
pulse 1/4
out+ pitch
in tempo
pulse 3/4
+ pitch
base 2
mul 330hz

This sequence will play a descending series of quarter notes spaced by an octave.

Wobble bass

in 135bpm
out b
in 55
mul 5
mul b
filt b

Siren (note similarity to wobble)

in 0.2hz
mul 440hz
+ 110hz
mul 2


in 120bpm
mul 8
gt 0.5
out b
in 5
gt 0.9
mul b

Note that the operator gt (greater than or equal) shapes the osc into a pulse wave where threshold is the operand and so sets the pulse width.

Basic Sample manipulations

in wavR
wav [name of wav file]
out dac

in mousex
filt 1hz
wav [name of wav file]
out dac

in wavR
mod 0.1
wav [name of wav file]
out dac

The mousex register supplies the relative X co-ordinate motion transmittted by the mouse. The second example simulates vinyl and the third controls the sample length. out dac is used here instead of mix assuming the sample is already pre-mixed.

Simple time-stretch algorithm

in 50hz
mul 0.25/50
mul 0.9
out b
in wavR
mul 0.1
+ b
wav [name of wav file]
out dac

The values 0.9 and 0.1 should sum to 1 to maintain original pitch. 0.25 is the frequency given by wavR for a sample length of 4 seconds

Basic reverb

in 0.3
lt 1/20
out b
in 440hz
mul b
+ c
tape 1
tap 0.1
+tap 0.3
+tap 0.07
mul 0.25
filt 1200hz
out c

The reverb begins at the + c line. the output is pushed onto the stack before being fed into tape. The feedback is from multiple taps which are attenuated by mul 0.3 before being fed back via the register c. The listing preceding the reverb generates a 440Hz sine wave gated by a pulse every 3⅓ seconds.

FM Bell

in 0.5
base E
+ -1
out b
in 280hz
mul 190hz
+ 105hz
mul b

Here, we use the base operator to shape the inverted ramp wave from osc into an exponential decay to control the amplitude of the bell. Feeding the output of the second osc into a third one results in FM synthesis, that is the output of one oscillator controls or modulates the frequency of the next.

UK dialing tone

in 1/6
lt 1/3
out b
in 480hz
out c
in 440hz
+ c
mul b

Although not perticularly musical, this simple necklace illustrates mixing two signals and gating them (turning on and off) with a third signal which is a pulse wave. The pulse function can be used instead to gate a signal by a variable width. slew 150 could be added before out b to smoothen the pulse for a less clicky sound.

Euclidean Rhythm

in 120bpm  
mul 1/4  
euclid 3,8,0  
decay 0.999 ◊  

Functions can have up to three operands separated by commas with no spaces. Refer to the function reference below for how many each one takes. The third argument of euclid is the phase offset of the internal oscilators.


The notation [a,b] is a closed interval, which means the numbers between a and b, including a and b.

List of operators

Operator Requires operand? Notes
in yes input from signal or fixed value
out yes output to signal
out+ yes add to signal
+ yes add previous result to operand, negate operand to subtract instead eg. + -1
sine no apply sine mathematical function
mod yes modulous operator
gt yes result is 1 if greater than or equal to operand, 0 otherwise
lt yes result is 1 if less than or equal to operand, 0 otherwise
mul yes multiply operator
abs no absolute value, all inputs become positive (removes negative sign)
tanh no hyperbolic tangent, useful for ‘soft clipping’
cvflt yes filter at fixed frequency of 100 hertz, requires unique register for internal calculation
clip no restrict input between symmetrical thresholds ±operand value. 0 is a special case resulting in thresholds of 0 and 1
noise no result is a pseudo-random series of numbers in range ( [-1, 1] * input )
pow yes result is operand raised to the power of input
base yes result is input raised to the power of operand
<sync yes receive sync pulse which zeros whatever is passed through. Operand adds phase offset on pulse
>sync yes send one sync pulse to all listings when input ≤ 0. Latches off until input > 0. More than one instance will result in undefined behaviour ◊
.>sync yes equivalent to >sync but will end listing and transfer, like out dac
nsync yes (not implemented) send one sync pulse to the listing whose index is given by the operand, triggered similarly to >sync
.nsync yes (not implemented) equivalent to nsync but will end listing and transfer, like out dac
push no move result to a stack
pop no take most recently pushed result from stack
tape yes record and playback from a rotating buffer, analogous to a tape loop. Operand is playback speed, record is at a constant rate. Use 1 for no change in pitch. At present can only have one instance of this operator per listing
tap yes result drawn from tape, operand is offset in seconds/milliseconds (use types)
+tap yes same as tap except added to previous result in listing
f2c no convert frequency to filter coefficient. Numbers less than than 0 will be multiplied by -1 (sign removed, become positive)
erase yes operand is number of operations to erase above the current in listing. For all use : erase. Currently, named signals invoked with ‘out’ in erased operations will be abandoned and remain unavailble, this does not affect correctness of the listing.
degrade yes add to a signal and draw result from a signal at random. Operand is a proportion amount. Will lead to unpredictable behaviour and eventual silence from that listing.
tempo yes (not implimented yet) ◊
pitch yes (not implimented yet) ◊
wav yes will play the corresponding sample of a loaded WAV file given by the operand. Expects an input in range [0, 1], values outside this range will wrap around this interval. See section below for more information
8bit yes quantises input to 8 bits of resolution (255 possible values). The Operand scales the input/output. 0 < input < 1 will have bigger quantisation steps and vice versa for input > 1.
level yes changes the output level of the listing at the index given by operand. The preceeding input sets the level. Level will persist after deletion. Capable of modulation up to 1200Hz, but because of this sudden large changes in level may produce clicks.
x yes alias of mul
* yes alias of x
from yes receives output of listing given by operand. If that listing is subsequently deleted it will receive from the one that takes its place. If the operand is greater than the number of listings it will wrap round (modulous)
sgn no outputs is 1 if the input is positive and -1 if negative
/ yes subtracts the operand from the input repeatedly until zero and outputs the number of subtractions as a fraction. AKA divide.
sub yes subtracts the operand from the input
setmix yes used internally for mix function
.level yes equivalent to level except will end input and launch listing
print no prints value of input to info display and passes through unchanged to next operation. Timing is a random point within an interval every 341.⅓ms
propa yes used in conjuction with index, adds multiple listings at once (not implimented yet) ◊
fma yes fused multiply add, the result of the input multiplied by the operand is stored in a special register fma (not implimented yet) ◊
List of commands (won’t appear in listing)
] no end function definition
: yes perform command: exit, erase, play, pause, fon, foff
fade yes changes fade out time after exit. Default is 325e-3 (unit is seconds, maximum 100)
del yes delete an entire compliled and running listing numbered by operand
index yes access index of listing
mute yes mute or un-mute listing at index given by operand. Muting won’t affect sync operations sent by a listing
unmute no un-mute all muted listings
solo yes solo listing at index given by operand. Solo-ing the same listing twice will have no effect
release yes set the release constant of the built in limiter. The limiter VCA envelope will decay by approximately 70dB in the operand time given in milliseconds. Default is 1s. Times of less than ~200ms may result in audible distortion or pumping. Times greater than ~2s will have a slow response to a decrease in level. The limiter has absolute peak detection (non-interpolated) and the attack (onset) is instantaneous. The decay curve is not stricly exponential as it has a slow onset to avoid distortion. Any listings that are much louder than the others will bring down the volume of all listings.
.mute yes equivalent to mute except will insert ‘out dac’ to launch listing
.del yes equivalent to del except will insert ‘out dac’ to launch listing. Used in effect to replace a listing
.solo yes equivalent to solo except will insert ‘out dac’ to launch listing
List of built-in functions
inv no invert a value between [0, 1], result equals 1/input
flip no turn a value between [0, 1] ‘upside down’, the input is flipped around y=½. Not suitable for negative values
tri no shape a value in range [0, 1] from saw/ramp to triangle. (mul 2, + -1, abs)
osc no ramp wave, (phase accumulator). Output in range [0,1]. Has DC offset of ½
saw no saw wave, descending ramp
mix yes output adjusted level to soundcard and end listing
s/h yes samples and holds input when operand crosses zero. Use ramp to supply operand. See ‘Sample and hold melody’ example above. A pulse will do track+hold, zero = track, any other number is hold.
dirac no result is a single pulse ◊
dist yes distortion, operand controls amount
sino no sine wave oscillator
filt yes 6dB per octave low-pass filter. Operand is cutoff frequency in Hertz
heat yes ‘bulb-element’ emulator (untested) ◊
cv2a no convert range [0, 1] to [-1, 1]
test yes output a sine test tone at the frequency of the given operand. Watch the volume as this is output at full scale!
decay yes will decay away to nothing from 1. 0.9997 is approx 20s, lower is quicker decay. Resets when input goes from 0 to 1
half yes like decay but accepts an operand in seconds that defines the ‘half-life’ of the decay. Input will override decay.
once no like osc but only completes one cycle. Operand will set upper limit and will reset when 0. Use 1 for one-off ramp
pulse yes pulse generator with duty cycle (pulse width) set by operand. Output is between 0 and 1, follow by cv2a for audio out
ramp no like osc but with an output suitable for audio, i.e. spans -1 to 1
posc yes like osc but will retrigger on a sync pulse. Operand sets phase offset
slew yes slew generator. Swings to the input at a rate given by operand. Try 150hz to reduce clicks on vca signals ( i.e. multiplying audio values)
T2 yes implements Chebyshev polynomial of the first kind. In plain english this means it will double the frequency of anything passed through it. Operand applies a level control, use 1 to pass through
zx no detects negative-going zero-crossing of input. A preceeding ramp will generate a single pulse of 1 at the end of its cycle.
lmap yes implements the Logistic Map. Iterates on zero-crossing of the input. Operand is the r value, suggested between 3 and 5. Preceed with ramp and follow with cv2a for audio output
euclid 3 outputs euclidean rhythms at the frequency given by input as a series of pulses. Eg. for (3,8) = “X..X..X.” the X will be 1 and the rests 0
exp no converts interval [0,1] to exponential.
dial no plays uk telephone dialling tone
dirac no outputs a single sample pulse when input goes from 0 to 1. Will trigger on first run of listing if input is 1
range 2 spreads input from 0 to ±1 across a range of values from the first operand to the second. Eg. range 220hz,440hz. If the second operand is smaller the range will be negative.
bd909 2 unfinished ‘909’ kick drum. first operand is decay and second is pitch.
List of pre-defined constants
ln2 natural logarithm of 2
ln3 ” ” 3
ln5 ” ” 5
E the mathematical constant e
E-1 1/(E-1), useful in base E,+ -1, mul E-1
Pi π, ratio of diameter to circumference of an ideal circle
Phi φ, the golden ratio ≈ (1+√5)/2
invSR 1 / SampleRate
SR SampleRate
Epsilon Epsilon, smallest possible number within syntə
wavR Frequency for wav playback at SampleRate, 0.25Hz at 4s sample time. Will need to be adjusted for samples with a different rate
semitone The twelfth root of 2, ratio of one semitone
List of pre-defined signals
dac signal represents output to soundcard. For use as out dac only
tempo not implemented yet
pitch not implemented yet
mousex value of mousepad X-coordinate
mousey value of mousepad Y-coordinate
butt1 value of left mouse button, 0 or 1
butt2 value of centre mouse button, 0 or 1
butt3 value of right mouse button, 0 or 1


Although theoretically possible, syntə is not primarily designed for making ‘normal’ music. A suggested use is the composition of waves, frequencies, shapes and combinations thereof to express a feeling or to dance to. The ideal aspect of syntə is using the full power of a modern compute to produce musical structures that would otherwise be difficult to materialise. The simple building blocks help gain insight into the workings and provide almost orthogonal flexibility. (If something is orthogonal it means the smallest number of parts to realise the full possibilities of a concept space.) Syntə enables detailed specification of sounds because it is built from small atomic operations. This gives a lot of freedom, which in turn requires some learning and practice. If you are new to synthesis you will need to learn that to, the program itself won’t teach you. However, in the opinion of the author the concepts and maths of audio are a delight to uncover and play with. Most of the time simple arithmetic is all that is required for a deeper understanding. Take small steps and be surprised at what you achieve in time. The language is intended to expand and suppliment the existing live-coding space to offer other possibilities and benefit the ecosystem as a whole.

Listings are asynchronous and will start immediately on submission to the sound engine. A >sync operation can be used to synchronise listings using posc, with any offset.

Sample Playback ◊

Any wav files placed in a folder named wavs will be loaded on startup. They need to be either 16 or 24bit resolution and either stereo or mono PCM format files. You will need to adjust the frequency of playback for wavs with non-standard sample rates. The nominal playback frequency is given by the pre-defined constant wavR which you can use for normal playback like so:

in wavR  
wav [name of wav file indicated above listings]  
out dac  

In the necklace above in sets the frequency of playback which is passed to osc to generate a rising ramp wave at that frequency. This ‘scans’ through the sample using the wav operator. Another way to look at it is that the output of osc, which repeats every 4 seconds, is ‘shaped’ into a sample waveform by wav. Any number between 0 and 1 will play the corresponding value of the sample at that point. Eg. 0.1 will play the sample at 400ms in. A continuous stream of values from osc will play the entire wav file. A decreasing series of numbers produced by, for example, osc, inv will play the sample backwards. inv flips the values of osc upside-down so the count down instead of up. Different frequencies fed to osc will change the speed of playback and adding a fixed number will offset the starting point of playback. So

in wavR  
+ 0.5  
wav [name]  
out dac  

will playback starting halfway through the sample. You don’t need to worry if the input value exceeds 1 as it will automatically wrap around so eg. 2.37 will become 0.37 and remain in range. As the default wav length is 4 seconds, 1 second of the wav corresponds to input of 0.25, 2s to 0.5 etc. This currently still applies to wavs shorter than 4 seconds, silence will played until the 4 seconds is up. ◊ By increasing the frequency and decreasing the amplitude of osc, a shorter section of the wav will be played.

wavR is a pre-defined constant which gives the playback speed for a sample recorded at the same rate as the internal sample rate, which is typically 48kHz. To adjust to playback speed for samples recorded at a different rate simply mul 44100/48000 for example, where the desired rate is 44.1kHz. This could be put more simply as mul 441/480, which is equivalent. To show this in a working necklace for clarity:

in wavR
mul 441/480
wav [name]
out dac

By design only the first 4 seconds of any wav file will be loaded. Two reasons for this are fast-loading times and to encourage creativity. You can manually edit any wav file in a free program such as Audacity to ensure the part you want to play if within the first 4 seconds, ideally starting at the beginning of the sample. Samples less then 4 seconds long are fine.

Later on, the intention is to provide more flexibility in synchronising other necklaces to wav files using len<wavname> signals. There is still plenty to be done in testing and possibly reworking parts of sample playback. ◊

Tape loop ◊

Each listing has a tape loop available which is accessed by the tape operator. The tape loop is a rolling buffer of 1 second in length. The output of tape is the ‘record head’, which is the input delayed by one second. For other delay times the loop can be accessed by the tap operator, where the fractional part of the operand sets the delay time. For example an operand of value 2.3 will give a tap at 300 milliseconds (0.3s). Multiple tap operators can be used. The use of multiple tape operators is undefined.

Arithmetic operations

An operand may be added in a listing of the form a/b or axb where a and b are valid numbers and the result is divison and multiplication respectively. For example: typing

in 2/3  

will result in

in 0.666666666...  

being added to the listing


Signals are the way of passing values around outside of the main flow through the necklace. Signals which are named and not just numbers can be referred to as registers, because they register a value. Usually the default value of a signal is 0. If you want it to begin as 1, add ‘ to the name. So a becomes 'a. Likewise you can use the double quotaion mark ” to have a default value of one half. If you want to be able to overwrite the value of a signal with out (more than once in a necklace), which is not normally possible, you can add ^ to the name. So a becomes ^a. You can use both of these special symbols, but the circumflex ^ must come first or it will be ignored.

Channels ◊

At present the output is in mono. Stereo output should be possible and is intended to be implemented in future. This will probably be via out L or out R, while out dac sends to bothleft and right channels. Please feel free to badger the author if you are keen for this to happen.

Synchronisation ◊

By default (and by design) when a listing is sent to the sound engine it starts immediately. The synchronisation operators >sync and \<sync can be used to co-ordinate listings to play in time with one another. First, for a rhythmic element use the phase synchronised oscillator posc instead of osc. This contains a \<sync operation which will reset the waveform when it receives a sync pulse. The posc operator requires a number betweeen 0 and 1 to offset the phase, you can use 0 to start with and experiment later. For reference 0.5 would result in a phase shift of 180°, 1 would be 360° etc. To send a sync pulse you can use >sync to synchronise all listings containing a posc or nsync to synchronise just one. For >sync you can either:

in -1

which will send one pulse. The . allows .>sync to end a listing and send it to the sound engine. The listings containing posc will remain in sync, even if you then delete this listing. Or, for periodic syncing you can:

in 120bpm

which will send pulses at the frequency given, in this case 1Hz. The saw function is similar to osc except the output spans between -1 and 1 which is necessary to trigger the pulse. Muting a sync listing will have no effect on the synchronisation, as it is send via a separate channel.

The proposed nsync (not implemented) works in the same way except that it requires an operand to determine which listing to send to. Any operands given that do not correspond to the index of a listing will be wrapped, that is to say the number of listings will be subtracted until in range. For a running necklace of index 2 this will sync it once:

in -1
.nsync 2

At present the behaviour of submitting more than one instance of >sync to the sound engine is undefined. This may change in future. ◊ The synchronisation is somewhat rudementary, a world away from DAW/midi sequencers, yet it has been designed to be raw and flexible in keeping with the Syntə philosophy. It also allows for a modicum of ‘musicianship’ as it is possible to submit listings in time with one another by hand (without sync) if you are that way inclined. Of course this is live coding which only intersects with music in general 🙂

Setting levels

The level operator is used to adjust the audio level of a running listing, like so: in 0.5 level 2 which would set listing two to half the available level, whcih is -6dB. This level-setting listing may be deleted straightaway and the level will persist once set. Although intended for basic mixing of listings, you may modulate the level at rates up to 1200Hz. This is an arbitary limit set to reduce clicks produced by immediate large changes in volume, while still allowing frequency modulation. You need to use .level to end a necklace.

Adding functions

If you find yourself reusing the same chunk of code multiple times, it is possible to define a named function which will instantiate that chunk of code. To begin, type the new name followed by [. Then type the listing as normal and at the end type ] (no operand) which will complete the function add. This function won’t be saved on exit but may be used as you wish during the current session. To permanently save a function which you feel will be useful in future type : fon before entering and it will be saved to the ‘functions.json’ file in the folder on exit from Syntə. To go back to ephemeral functions (useful for experimentation) type : foff. You may overwrite functions by typing in the same name. If the function has no operands type the “!” character before the name with no spaces to overwrite. Eg. !blah [ The ability to make functions like this makes the language extensible, which means you are able to extend the language beyond what is written in this document. One of the project aims is to build up a library of abstractions in this way to make performance easier for beginners. However there is a limit to this, as just typing ‘music’ and stopping there would be quite boring! An abstraction means wrapping up a bit of code into something simple to make it easier to use, for example the term ‘global apartheid’ is an abstraction of a system and history that involves many many processes, interconnections, organisations, trade-misinvoicing etc.

info.go and listing.go

info.go is intended to run alongside Syntə to display useful information and error messages. The layout is as follows:

Syntə info *press enter to exit*		0s		<-- elapsed running time in seconds

					Load: 0.00		<-- If load approaches 1, glitches or dropouts may occur in the audio

								<-- info and error messages will appear here

								<-- list of active registers in current listing input will appear here
	0.00	|||||||			|			<-- peak audio meter, will display 'GR' if limiting takes place on the output.
      Mouse-X: 0.0000		Mouse-Y: 0.0000			<-- value of mouse X and Y

listing.go displays the currently running necklaces. Any that are muted will show in grey italics. The functions within a listing are ‘unrolled’, that is to say they are shown in terms of their atomic operations.

Hot tips

  • use slew 150 to smoothen a signal used for modulating volume with mul
  • trim samples of music to be an exact number of bars (look for repetion in the wavform of around ~2 seconds)
  • use the “!” type to use a number outside of the expected range if you need to. This is not a factorial operation
  • had a moment of inspiration and can’t remember what you did? All listings accepted by the sound engine are saved by timestamp to the recordings folder in json format
  • other tips tba… ◊

Performing with Syntə

To prepare the audio system:

  1. Set up the equipment and verify sound from Syntə is reaching the speakers
  2. Turn the system down to zero
  3. Type test 2000hz
  4. Increase volume until an acceptable level You will probably need to go above this level for perfomance once you have an audience in the room and are playing full range sounds, but this gives a good starting point that you probably don’t want to go too far over. Please ensure the audience has access to hearing protection if the system is capable of more than 85dBSPL. To specifically test the bass level and evenness try:

in 10s
mul 4
base 2
mul 20hz
out dac

If the bass level fluctuates a lot at different frequencies and at different listing positions you should consider some using some bass trapping. This is beacause sound reflections between the walls will interfere and cancel out.

The Sound Engine

Although it is not necessary to know how the sound engine works to perform or play with Syntə, it can be helpful to learn more about it so we’ll give a brief outline here. When a listing is send to the sound engine an internal copy is generated and added to the sequence of listings. The sound engine takes each listing in turn from 0 onwards and runs through it once to produce the value of the next sample, then it moves on to the next listing. Once it has computed all the listings the resulting samples are summed together and sent through the built-in limiter to ensure no loud surprises and the peak loudness of the output is also sent to the info display. The resulting signal is converted to the correct format and sent to the soundcard in your computer, after which the whole process repeats. The terms ‘listing’ and ‘necklace’ are interchangeable. Necklace also illustrates the point that the listing is computed in a continuous loop. Although usually the result is computed from fresh each time, unless you use the from operator. All values in the sound engine are represented by 64-bit floating point numbers which have a nominal range within Syntə of between -1 and 1 inclusive. At the end of each cycle of the sound engine this is converted to a 32 bit number to be sent to the soundcard. Within a listing a value can be anywhere in the range of the 64-bit float (approx ±1.8×10^308 with an precision of about 15 decimal places.) These numbers will be limited or clipped to ±1 before audio conversion. The samples from each listing are added together and when there are more than four listings the sum is divided by the number of listings for unity gain. For 1, 2, 3, or 4 listings the sun is always divided by four. The result is then passed through a high-pass filter before the limiter. This is to remove any DC offsets, which means a consistent average signal other than 0, or another way to think of it is attentuating (reducing) frequencies below 4.6Hz. If my calculations are correct, any DC signal will fade below -120dB after half a second. The limiter reduces the level of audio above a peak value of 1 to avoid the possibility of clipping which would produce distortion. The detection algorithm of the limiter is more sensitive to higher frequencies, it expects audio to have a spectrum approximately equivalent to ‘pink noise’. Between the limiter and the clipping stage before conversion, what is known as dither is applied to the signal. This is a tiny amount of noise to avoid rounding errors, but is probably overkill. before the dither any envelopes associated with pausing or exiting are applied. These reduce the chances of pops or clicks. The whole main loop of the sound engine is timed to produce the ‘load’ value that shows how much work it is doing to create each sample for the soundcard. See info display section above.

A note on licensing

While for now the work is unlicensed, it is safe to assume that:

  • Running the software in person (not via a network) for the purposes of perfromance or education is unlikely to lead to any negative consequences.

  • Using, modifying, or distributing the software in binary or source code form for commercial purposes will never be permitted.

Suggestions, including improvements to the code are welcome. We are unable to accept code contributions directly until the project has been licenced. If you would like to fork the code for more radical changes then we wholeheartly suggest you instead write a unique live coding platform from scratch. The file is of an order of only two thousand lines of code after all, and the learning experience will be invaluable to you.

Influences and other live-coding environments

The term live-coding is generally taken to mean on-the-fly composition of algorithmically generated audio or graphical artifacts using software. We recommend exploring TidalCycles and Pure Data in particular. If you are more visual/graphical oriented there are live coding languages/environments for that space too. Many influences and inspirations have been drawn upon either deliberately or subconciously in the creation of Syntə. They include assembly language, Forth, the live-coding laguages mentioned above, modular synthesis, Musique concrète, Go itself (the language Syntə is implemented in) and unix-like systems in general – primarily FreeBSD which was the first OS to host Syntə, both for development and execution.

Future possiblities for Syntə

It would be great if translations of this document and the language itself become available in future, get in touch if you can help with this. A live distro with Syntə included could be useful for users of other operating systems to boot from. A YouTube channel to stream live demonstrations/performances is planned ◊. Another possiblity for the code is integration with a terminal library such as tcell or bubbletea for a slicker UI. Also, the Sound Engine could be adapted to a ‘headless’ mode for use with an editor such as Atom. This option hasn’t been pursued for now as it is helpful to provide immediate feedback via info.go with each operation entered. Future additions include providing for audio input and use of other sound drivers than OSS/ALSA. A browser-based interface is another option, which would make the program extremely portable. This could be as simple as redirecting stdout to a local websocket server. You may consider the current implementation of Syntə to be a prototype. At present most of the basic features have been taken care of, and as the language is tried out by different people, changes and adjustments that will tidy up and make performing easier may become apparent. The core aims of fun and accessibility won’t change.

General principles of live coding performance

These are, as the author understands it (opinions may differ):

  • Display your code to the audience

  • Prefer fresh code over copy-pasta

  • Improvisation and exploration are key

  • An accessible and welcoming space is essential

  • Share and mentor where possible

  • Mistakes and errors are embraced and honoured

To do: (for this document)

how to modulate pitch and volume, including ratios and exponents feedback and what it means filtering delay and reverb pause/play mouse control tempo and pitch (not fully implemented yet) euclidian rhythms fractal synthesis (not implemented yet) finite recursion (not implemented yet) mute and solo


View Github