A Rust port of the Sonant 4K synth with streaming support.
Sonant (C) 2008-2009 Jake Taylor [ Ferris / Youth Uprising ]
What is it?
A tiny synthesizer written for 4K intros. It is capable of producing high quality audio with very little code and instrument data. Song files are just over 3KB, but can also be customized to reduce the number of instrument tracks or patterns if you have a tighter size budget.
sonant::Synth type is implemented as an iterator, which makes it ideal for producing realtime audio streams with very little memory overhead; about 6.2 KB for the song data, and another 2.5 KB for buffering note frequencies. It was originally written to target Nintendo 64, which has a baseline of 4 MB of system memory!
Unfortunately, it's too slow to run on the N64's 93 MHz CPU. It would probably work on the RCP, e.g. by computing 8 samples at a time on the vector unit. But that would require porting the sample generators to use 16-bit fixed point numbers. Then there's also the problem that rustc cannot target RCP. Oh well!
How does it work?
Flippin' maths and magics! I have no idea. Synthesizers are weird and alien to me, but they make really pretty ear-candy.
Each song has eight instrument tracks, and each instrument has two oscillators. The oscillators work together (or adversarially canceling each other, if you like) to vary the instrument frequencies. The "personality" of the instrument is provided by one of four waveforms:
Triangle. The oscillators' frequencies modulate these basic waveforms to produce the final sounds.
In addition to the primary oscillators, each instrument also has its own LFO, which is what makes that slow pitch-bending that you hear all the time in electronic music.
Finally, each instrument also has it own effects channel, which can do
Notch filtering. The effects also provide simple resonance, delay (echo), and panning.
The rest of the song structure is pretty standard for tracked tunes; Each instrument can have up to 10 patterns. And any pattern can be referenced from a 48-element sequence. Each pattern itself contains 32 notes.
Delay effects are implemented as extra notes, which greatly reduces the memory footprint. The original implementation uses over 42 MB of memory to maintain the delay buffers. I made the tradeoff to pay for better memory efficiency by recomputing all of the delayed samples as they are needed.
See the Sonant manual (bundled with the original release archive on Pouët) if you would like to learn more about the synth, tracker, or song format.
How to use it?
player example for some code that loads and plays a
cargo run --release --example player -- ./examples/poseidon.snt
The original synthesizer doesn't have many limitations beyond what the
.snt format is capable of storing. The iterator-based implementation of this port does come with a few restrictions, though. For example, only up to 8 overlapping notes are able to be played simultaneously for each instrument track.
sonant-tool is capable of producing
.snt files which require up to 100 overlapping notes per instrument track, but this is only true in the most extreme possible case. The
.snt format itself is theoretically able to require up to 1,536 overlapping notes!
Songs which use a lot of delay effects on the instruments will more quickly hit the overlapping note limits. If you need to support more overlapping notes, you can simply increase the value in
consts.rs; any value up to 32 will work without any other changes.
Due to the way the delayed notes work, the length of quarter notes cannot be an odd number of samples. This would cause the length of eighth notes to be a fractional number, and would complicate the process of "finding notes in the past". To resolve the conflict, the length of quarter notes is adjusted to an even number by "rounding down" to the nearest even number. This has a small impact on playback duration; a four-minute song will be about 1 second shorter than it would as rendered by other players.
Sonant generates samples in reverse order. We have to generate samples chronologically. This shifts the phase of the waveform for individual notes arbitrarily (it depends on note length, envelope, and the nondeterministic LFO). The differences are too subtle for humans to distinguish, but it is worth mentioning.