Designing synthesis systems

Whenever process is too complex

when a file needs to be the final product

when the output is analysis data from a file

Because process is too complex

or because files need to be written as fast as possible

instead of a single loop to process audio

a small loop must be called time and time again

it's a good idea to always write this way as this mechanism is also suitable for non-RT rendering

Callback is most efficient as only buffers requested by the hardware are filled

but for the same reason it is more demanding of system resources

Blocking methods write to an intermediate FIFO queue which is shared between the synthesis thread and the hardware thread

This generally requires larger buffers and latencies

block size can be equal to control block size or not

No blocking operations

algorithms with constant time preferred.

if low latency is not required these requirements can be relaxed, particularly if audio is not being calculated directly in the audio callback but to an intermediate buffer on a separate thread.

a connected list of nodes that represent the audio processes

The graph is traversed to obtain the audio output

processes are applied one after the other.

The graph may contain several instances of the same "algorithm" for various simultaneous notes that are playing

Unit generators are building blocks for audio synthesis

Devised by M. Matthews for Music III in 1960

Allows the flexibility of interconnecting an indefinite number of them in any order

Generally more efficient as some overhead is removed

But of course less flexible

e.g. Gamma, CSL

"notes" can be added to the graph while the graph is being rendered

Usually a set of instruments or SynthDefs are available to instantiate on the graph

e.g. Csound, SC, PD

From the bottom

From the top

Unit generator must have some sort of unified interface to be able to be stacked in a graph

i.e. they must share some functions that the graph will call to get their output

This can be done easily in object oriented languages thorugh inheritance

in C, the first member of a struct can be used to simulate inheritance through pointer casting

Apart from the process function, init and destroy functions are generally needed

Parameters

Unit generators are usually grouped together in instruments/synthdefs

instruments can be allocated to create running instances

In PD, Chuckk or Max/MSP, ugens can be placed directly in the rendering graph, not as part of instruments.

tickers

Control block

multi-rate