Oidos is a software synthesizer, based on additive synthesis, for making music for very small executables, such as 4 and 8 kilobyte intros.
You can follow the devopment on GitHub. For discussion, visit the Pouet forum.
The synth has two parts: a VST instrument, Oidos, and an accompanying VST effect, OidosReverb. The VSTs can in principle be used in any DAW, but the toolchain for using the music in an executable assumes that the music is made using Renoise.
Each instrument uses its own Oidos VST instance. The OidosReverb effect VST can be added as a Track DSP to add a reverb effect to some of the tracks.
The synth is quite computationally heavy, especially when the modes and fat parameters are set to high values. The VST internally caches the sound produced by each tone, so as it gets "warmed up" on particular instruments, it gets less heavy to work with. You will sometimes hear some stuttering in the sound the first time a tone is played. It can be useful to disable "overload prevention" in the Renoise settings.
To be able to convert your music into executable form, you must adhere to these guidelines:
- You can use as many tracks, and as many note columns within each track, as you like.
- Each note column must contain notes from only one instrument. You can use each instrument in as many tracks and columns as you like.
- You can use per-note velocity, which will scale the volume of individual notes.
- You can not use the panning, delay or effect columns.
- You can use Send devices, but only in "Mute Source" mode.
- You can adjust volume and panning using Instrument Volume, Track Volume/Panning, Mixer Volume/Panning, Send Volume/Panning and Master Volume/Panning. However, all tracks using the same instrument must have the same volume and panning. This is most easily accomplished by grouping all notes played using the same instrument into one or more note columns within the same track.
- You can only use one OidosReverb instance. This is typically placed on a Send track, with some tracks routed to it. For each instrument, either all or none of the note columns using that instrument can have reverb.
- You can use group tracks, but only for visual grouping. You can not use volume, panning, Send devices or reverb on group tracks.
- You can use the pattern sequence matrix to selectively mute tracks at certain pattern positions.
- Globally muted tracks or note columns will not be included. Solo state is ignored.
The OidosConvert program in the convert directory will convert a Renoise
song using Oidos and OidosReverb into an assembly source file to be
included with the supplied player source. See the oidos.h
file for documentation on how to invoke the Oidos player.
Run the converter from the commandline with input and output file names, like this:
OidosConvert music.xrns music.asm
If your terminal supports ANSI escape codes and you want some nice colors for enhanced readability, use:
OidosConvert -ansi music.xrns music.asm
Pay close attention to the output from the converter, as it will tell you if it encountered an error along the way (for instance if one of the guidelines are violated).
If you are only interested in a stand-alone executable that just plays the
music (for instance for an executable music compo), there is a complete
setup for this in the easy_exe directory. It also produces a
WAV writer, as required by many executable music compo rules.
An important part of the workflow when producing music for a size-limited
executable is optimizing the size and computation time requirements of the
music. The OidosConvert program outputs some statistics about the music
which can be used to guide this process:
Memory: The memory needed to store precomputed sound for this instrument. For each tone an instrument is played with, the sound is precomputed at the length of the longest note played by the instrument. Thus, the memory requirement is proportional to the product of these 2 factors:
- The number of different tones the instrument is played with
- The longest note played by the instrument
The memory for precomputed sound is reused for every instrument. The maximum memory used for all instruments is printed at the end.
Burden: The total computation time requirements of all tracks using this instrument. The time requirement is a product of these 4 factors:
- The value of the modes parameter
- The value of the fat parameter
- The number of different tones the instrument is played with
- The longest note played by the instrument
The total burden for all instruments is printed at the end, along with an estimate of the real time for a reasonably fast CPU.
Tones: Lists all the tones the instrument is played with. The number after the colon indicates how many times in the song the instrument is played with that tone.
Velocities: Lists all the velocities the instrument is played with. The velocity values are automatically quantized to the largest power of two dividing all used values (with 7F treated as 80). Sticking to more "round" values will reduce the number of bits required to represent each note velocity. The number after the colon indicates how many times in the song the instrument is played with that velocity.
Lengths: Lists all the lengths (distance from each note until the next note or OFF) in this column, with the number after the colon indicating how many times that length occurs in the column. If all notes in a column have the same length, a more compact representation of the track is used, omitting all OFFs.
Notes: Lists the tone/velocity combinations used in the column, with the number after the colon indicating how many times that combination occurs in the column. Notes are represented as indices into a list of these combinations, so reducing the number of combinations will typically reduce the size of the music.
Using reverb will add around 100 bytes to the compressed size for the reverb code and parameters. Using panning will usually add some 10-30 bytes depending on the number of instruments.
Also be sure to quantize all parameters, as described below.
Oidos is an additive synthesizer, which means it produces sound by adding together a large number of sine waves, known as partials. The frequencies and amplitudes of these sine waves, along with their variation over time, determines the character of the sound. All of this is controlled by the VST parameters:
Random seed for all random choices in the synth. Changing this will often change the sound dramatically, even with the same values for the other parameters. Experimentation is the key here.
The partials are grouped into a number of modes - characteristic frequencies of the sound. This parameter specifies the number of modes.
Each mode contains a number of partials, grouped around the mode's center frequency. This parameter specifies the number of partials per mode.
Controls how spread out the frequencies are of the partials belonging to the same mode.
Controls the distribution of the center frequencies of the modes. The frequencies are randomly distributed between the base frequency (the played key) and this many semitones above the base frequency.
Controls how the amplitude of a mode depends on its frequency. With low sharpness, the amplitudes fall off strongly, resulting in a soft sound. With high sharpness, the amplitudes fall off less, or even rise with frequency, resulting in a sharp sound.
If all frequencies in a sound are close to overtones (integer multiples) of the base frequency, the sound is perceived as harmonious. Harmonious instruments are used for the tonal parts of music. Disharmonious instruments are for instance drums, which don't have a specific tone.
The harmonicity parameter pulls the mode center frequencies towards (or pushes them away from) overtones of the base frequency in order to make the sound more or less harmonious.
The decaylow and decayhigh parameters control how quickly the amplitudes of modes fall off over time. They specify the falloff rate for low and high frequencies, respectively. The falloff of each frequency is determined by interpolating between these two values.
A filter is applied to the partials before summing them. The filterlow and filterhigh parameters specify the low and high limits of this filter, relative to the base frequency of the note. Frequencies outside these limits are discarded or attenuated. The fslopelow and fslopehigh parameters specify the sizes of the sloped regions at the filter limits, i.e. the frequencies which are (increasingly) attenuated before the filter cuts off completely. The fsweeplow and fsweephigh parameters specify the movement of the filter limits over time.
A non-linear distortion is applied to the summed partials. The gain parameter controls the strength of this distortion.
Specifies the attack time of the sound, i.e. the time before the sound reaches its full volume.
Specifies the release time of the sound, i.e. the time before the sound reaches zero volume after the note is released.
All parameters beginning with q are quantization parameters. These parameters round off the internal floating point representations of the parameters to "nicer" values which compress better, thereby reducing the compressed size of the parameter data.
When finalizing a piece of music, go through all the quantization parameters of all the instruments and pull them up as high as you can without ruining the sound.
The quantization parameters will show the quantized values of the quantized parameters, and so will the parameters themselves. You can adjust a parameter which has been quantized, at which point it will jump between the values allowed by the quantization.
The included reverb effect is a simple, "strength in numbers" reverb, which simply consists of a large number of filtered feedback delays.
The parameters are:
The strength of the reverb.
Panning of the reverb. Can be used to center the reverb of a non-centered reverb source.
Each feedback delay making up the reverb will have a feedback delay length randomly distributed between delaymin and delaymax. Additionally, the whole reverb will be delayed by delayadd.
Controls how quickly the reverb dies out.
Filters the sound prior to the reverb. Frequencies below filterlow or above filterhigh will be attenuated.
Dampens the sound as it goes around the feedback delays. Frequencies below dampenlow or above dampenhigh will be increasingly attenuated as the reverb progresses.
Number of feedback delays making up the reverb. The more delays, the smoother the reverb.
Random seed for the random delay lengths. Some seeds cause the delays to interfere with each other, resulting in faint echos and irregularities in the reverb. Experiment with the seed to finetune the reverb.
All parameters beginning with q are quantization parameters, working the same way as described for the synth above.