The Synthesis ToolKit in C++ (STK)
By Perry R. Cook and Gary P. Scavone, 1995--2019.
This distribution of the Synthesis ToolKit in C++ (STK) contains the following:
include: STK class header files
src: STK class source files
rawwaves: STK audio files (1-channel, 16-bit, big-endian)
doc: STK documentation
projects: example STK projects and programs
For compiling and installing STK, see the INSTALL.md file in this directory.
- System Requirements
- What's New (and not so new)
- Legal and Ethical
- Further Reading
- Perry's Notes From the Original Distribution
The Synthesis ToolKit in C++ (STK) is a set of open source audio signal processing and algorithmic synthesis classes written in the C++ programming language. STK was designed to facilitate rapid development of music synthesis and audio processing software, with an emphasis on cross-platform functionality, realtime control, ease of use, and educational example code. The Synthesis ToolKit is extremely portable (most classes are platform-independent C++ code), and it's completely user-extensible (all source included, no unusual libraries, and no hidden drivers). We like to think that this increases the chances that our programs will still work in another 5-10 years. STK currently runs with "realtime" support (audio and MIDI) on Linux, Macintosh OS X, and Windows computer platforms. Generic, non-realtime support has been tested under NeXTStep, Sun, and other platforms and should work with any standard C++ compiler.
The only classes of the Synthesis ToolKit that are platform-dependent concern sockets, threads, mutexes, and real-time audio and MIDI input and output. The interface for MIDI input and the simple Tcl/Tk graphical user interfaces (GUIs) provided is the same, so it's easy to experiment in real time using either the GUIs or MIDI. The Synthesis ToolKit can generate simultaneous SND (AU), WAV, AIFF, and MAT-file output soundfile formats (as well as realtime sound output), so you can view your results using one of a large variety of sound/signal analysis tools already available (e.g. Snd, Cool Edit, Matlab).
The Synthesis Toolkit is not one particular program. Rather, it is a set of C++ classes that you can use to create your own programs. A few example applications are provided to demonstrate some of the ways to use the classes. If you have specific needs, you will probably have to either modify the example programs or write a new program altogether. Further, the example programs don't have a fancy GUI wrapper. If you feel the need to have a "drag and drop" graphical patching GUI, you probably don't want to use the ToolKit. Spending hundreds of hours making platform-dependent graphics code would go against one of the fundamental design goals of the ToolKit - platform independence.
For those instances where a simple GUI with sliders and buttons is helpful, we use Tcl/Tk (https://www.tcl.tk/) which is freely distributed for all the supported ToolKit platforms. A number of Tcl/Tk GUI scripts are distributed with the ToolKit release. For control, the Synthesis Toolkit uses raw MIDI (on supported platforms), and SKINI (Synthesis ToolKit Instrument Network Interface, a MIDI-like text message synthesis control format).
See the individual README's (e.g. README-linux) in the /doc directory for platform specific information and system requirements. In general, you will use the configure script to create Makefiles on unix platforms (and MinGW) or the VS2017 workspace files to compile the example programs. To use the Tcl/Tk GUIs, you will need Tcl/Tk version 8.0 or higher.
WHAT'S NEW (AND NOT SO NEW)
Despite being available in one form or another since 1996, we still consider STK to be alpha software. We attempt to maintain backward compatability but changes are sometimes made in an effort to improve the overall design or performance of the software. Please read the "Release Notes" in the /doc directory to see what has changed since the last release.
A new StkFrames class has been created to facilitate the handling and passing of multichannel, vectorized audio data. All STK classes have been updated to include tick() functions that accept StkFrames arguments.
The control message handling scheme has been simplified greatly through the use of the Messager class. It is now possible to have access to simultaneous piped, socketed, and/or MIDI input control messages. In most cases, this should eliminate the use of the Md2Skini program.
Realtime audio input capabilities were added to STK with release 3.0, though the behavior of such is very hardware dependent. Under Linux and Macintosh OS-X, audio input and output are possible with very low latency. Using the Windows DirectSound API, minimum dependable output sound latency seems to be around 20 milliseconds or so, while input sound latency is generally higher. Performance with the ASIO audio API on Windows provides much better performance.
As mentioned above, it is possible to record the audio ouput of an STK program to .snd, .wav, .raw, .aif, and .mat (Matlab MAT-file) output file types. Though somewhat obsolete, the program Md2Skini can be used to write SKINI scorefiles from realtime MIDI input. Finally, STK should compile with non-realtime functionality on any platform with a generic C++ compiler.
For those who wish to make a library from the core STK classes, the configure script generates a Makefile in the src directory that will accomplish that.
You probably already guessed this, but just to be sure, we don't guarantee anything works. :-) It's free ... what do you expect? If you find a bug, please let us know and we'll try to correct it. You can also make suggestions, but again, no guarantees. Send email to the mail list.
LEGAL AND ETHICAL
This software was designed and created to be made publicly available for free, primarily for academic purposes, so if you use it, pass it on with this documentation, and for free.
If you make a million dollars with it, it would be nice if you would share. If you make compositions with it, put us in the program notes.
Some of the concepts are covered by various patents, some known to us and likely others which are unknown. Many of the ones known to us are administered by the Stanford Office of Technology and Licensing.
The good news is that large hunks of the techniques used here are public domain. To avoid subtle legal issues, we'll not state what's freely useable here, but we'll try to note within the various classes where certain things are likely to be protected by patents.
For complete documentation on this ToolKit, the classes, etc., see the doc directory of the distribution or surf to http://ccrma.stanford.edu/software/stk/. Also check the platform specific README's for specific system requirements.
PERRY'S NOTES FROM THE ORIGINAL DISTRIBUTION
This whole world was created with no particular hardware in mind. These examples are intended to be tutorial in nature, as a platform for the continuation of my research, and as a possible starting point for a software synthesis system. The basic motivation was to create the necessary unit generators to do the synthesis, processing, and control that I want to do and teach about. Little thought for optimization was given and therefore improvements, especially speed enhancements, should be possible with these classes. It was written with some basic concepts in mind about how to let compilers optimize.
Your question at this point might be, "But Perry, with CMix, CMusic, CSound, CShells, CMonkeys, etc. already cluttering the landscape, why a new set of stupid C functions for music synthesis and processing?" The answers lie below.
I needed to port many of the things I've done into something which is generic enough to port further to different machines.
I really plan to document this stuff, so that you don't have to be me to figure out what's going on. (I'll probably be sorry I said this in a couple of years, when even I can't figure out what I was thinking.)
The classic difficulties most people have in trying to implement physical models are:
A) They have trouble understanding the papers, and/or in turning the theory into practice.
B) The Physical Model instruments are a pain to get to oscillate, and coming up with stable and meaningful parameter values is required to get the models to work at all.
This set of C++ unit generators and instruments might help to diminish the scores of emails I get asking what to do with those block diagrams I put in my papers.
I wanted to try some new stuff with modal synthesis, and implement some classic FM patches as well.
I wanted to reimplement, and newly implement more of the intelligent and physical performer models I've talked about in some of my papers. But I wanted to do it in a portable way, and in such a way that I can hook up modules quickly. I also wanted to make these instruments connectable to such player objects, so folks like Brad Garton who really think a lot about the players can connect them to my instruments, a lot about which I think.
More rationalizations to follow . . .