woyten / tune Goto Github PK

It should be possible to describe all currently possible effects as deserializable objects.

• Monophonic pitch-bend tuning
• Polyphonic pitch-bend tuning

It is not guaranteed that this will actually improve the performance.

Apply vibrato when modulation is non-zero.

Idea: Add an enum variant LfSoruce::Controller which reads the specified controller's current value.

• A Channel could be used to track the controllers' current state.

Add a tune live command to tune-cli.

• Just-in-time octave-based tuning
• Ahead-of-time octave-based tuning

I really can't figure out what the arguments to tune scale rank2 are. Can the documentation be extended a little?

Also the generation of tuning files for a synthesizer are unclear. I can not get tun to produce a file.

Move the live retuning functions from tune-cli to tune and make them accessible in microwave.

• All AOT live retuning functions available in microwave
• JIT live retuning for continuous mode Fluid synth
• All JIT live retuning functions available in microwave

Linux Mint 20.2

cargo b
error: failed to parse manifest at `~/src/tune/Cargo.toml`

Caused by:
  feature `edition2021` is required

  consider adding `cargo-features = ["edition2021"]` to the manifest

I'm sure this is technically a violation of the spec, which I'd guess is why this check is in the builder, but all the tools I use (e.g. ODDSound MTS-ESP) are happy to generate scales that violate this. For example. I realize this check might be helpful for manually building scales, but I'm wondering if there's a way we can disable it for import.

• Tune MIDI channel via Single Note Tuning
• Apply ChannelTuner

Current situation:

• LfSource acts as a specification and as a state object at the same time
• When using LfSourceExpr::Envelope the envelope needs to be looked up in a HashMap which seems undesirable in terms of performance

Desired situation:

• In the same way that StageSpec and WaveformSpec are separated from Stage and Waveform, LfSource should be split into a spec and state object. Done in 6e9b0fd.
• Envelopes should be looked up once on waveform creation and be stored in the state object. Done in 0a0b4b7.

• Retrieve keyboard layouts (Meantone / Porcupine) for any scale
• Render a hexagonal isoorphic keyboard

It has been observed that some waveforms aren't centered around the 0 baseline. A DC filter could be used to remove this offset.

Reference: https://www.musicdsp.org/en/latest/Filters/135-dc-filter.html

• Basic types
  • Pitches and ratios
  • Notes and keys
  • Ordered and unordered scales with inverse search
  • Keyboard mappings
  • API/Documentation quality
• Scala interop
  • SCL import/export
  • KBM import/export
  • API/Documentation quality
• MTS support
  • Scale/Octave Tuning
  • Single Note Tuning Change
  • RPN messages
  • Less common variants (with bank select / 2-byte format)
  • API/documentation quality
  • (optional) MTS parsing
• Ahead-of-time tuner
  • Full-keyboard based
  • Octave based
  • Channel based
  • API/documentation quality
• Just-in-time tuner
  • Full-keyboard based
  • Octave based
  • Channel based
  • API/documentation quality
• Adaptive tuner
• MIDI
  • Serialize/deserialize channel messages
  • API/documentation quality
• Generalized keyboard layouts and notation
  • Basixc Val and Comma types
  • PerGens
  • MOSes
  • Derive meantone and porcupine keyboard layout and notation and from PerGen and MOS properties
  • Support more temperament layouts based on actual commas being eliminated
  • API/documentation quality

Running microwave fails because it's unable to open a graphics device

[INFO] Loading waveforms file `waveforms.yml`
thread 'main' panicked at 'called `Result::unwrap()` on an `Err` value: NoAvailableAdapter', microwave/src/main.rs:451:10
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
zsh: segmentation fault  microwave run

It looks like this is a known issue. Vulkan assumes a reasonably powerful graphics card. I have

$ lspci | grep VGA    
00:02.0 VGA compatible controller: Intel Corporation Core Processor Integrated Graphics Controller (rev 12)

A headless mode should at least allow the synth to run

Mint Uma 20.2 using PipeWire 0.3.39

./microwave run --midi-in "QuNexus MIDI 1" connects to the device but does not appear to release the note when the key is no longer being pressed

Here are 3 examples, using keyboard and mouse which work well, and then midi which leaves the residuals

• Finalize layout matching concept (Val → PerGen → Mos)
• Generate colors based on obtained layout
• Send color schema to Lumatone

Add a keyboard mapping for ANSI keyboards s.t. adjacent keys map to adjacent virtual key coordinates.

% tune scale 440 equal 31                                                                                                                                                                                                                      
thread 'main' panicked at 'Ratio must be finite and positive but was 0', /home/suhr/.cargo/registry/src/github.com-1ecc6299db9ec823/tune-0.10.0/src/ratio.rs:49:9
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

There's a handy crate for error handling: https://docs.rs/anyhow

Render audio samples when playing a note.
Idea: Create an enum variant StageSpec::Sample which renders an audio file to a target buffer.

• Define an execution model / How can different CLI calls interact with each other? Examples
  • REPL

    $ tune
    > scl steps 1:12:2
    > kbm D4
    > connect QSynth
    > mts
    > exit

  • Pipeline

    $ tune scl steps 1:12:2 | tune kbm D4 | tune mts Qsynth

  • Endless args

    $ tune scl steps 1:12:2 kbm D4 mts Qsynth

• What calls should be supported?

Example scale which doesn't work at the moment:

100.0
200.0 = overshoots the period
150.0 = last item / period

Just waiting until nannou-org/nannou#657 is solved might be a solution.

• Implement the full kbm model
• Import kbm files
• Use full kbm model in tune-cli
• Use full kbm model in microwave

Use a binary search on a Stern-Brocot tree to find the nearest rational number approximation.

The current performance does not seem to be a a problem. microwave can render 800 Sin stages and 1600 Sawtooth stages simultaneously on a single Core i7-6700HQ CPU @ 2.60GHz. There is still room for improvement by rendering the Sin function to a wavetable (#28) and by using all CPU cores.

However, thinking about performance without measuring it is a problem. Therefore, I would like to add some benchmark tests that confirm or disprove performance considerations.

Vision:

• Add a bench subcommand which runs some performance test (e.g. measures the time to render 100.000 Sin stages into a buffer)
• Create an independent github action which will build the microwave executable for all commits since the latest release tag in --release mode
• Run microwave bench for all builds multiple times and create a benchmark report

Starting microwave on a Windows PC fails with a StreamConfigNotSupported error.

Cause: Setting StreamConfig::buffer_size to BufferSize::Fixed(<value>) is not allowed.

Fix: Set StreamConfig::buffer_size to BufferSize::Default when SupportedStreamConfig::buffer_size is SupportedBufferSize::Unknown.

Ideas:

• Low Pass
• High Pass
• Resonance
• Spatial Delay
• Chorus
• Reverberation
• Phaser
• Flanger
• Time-dependent amplitude / Tremolo
• Time-dependent filter
• Time-dependent panning
• Leslie effect

More ideas: https://rickyhan.com/jekyll/update/2018/02/06/rust-guitar-pedal-effects-dsp.html

Retune played notes in real time s.t. their frequency ratios get as close to JI as possible.

Usage: The user provides a list of (pure) intervals some of which can be left empty.

Example: x, 9/8, 6/5, ... means minor seconds have no preference, major seconds prefer to become 9/8, minor seconds prefer to become 6/5, etc.

Idea: Optimize two energies

• The sum over all intervals of the squared difference between the actual interval and the just interval
• The sum over all pitches of the squared difference between the actual pitch and the original pitch

Enhancements: Provide a weight for each ratio.

Couple the rotary effect to the expression controller (overrideable via command line parameter)

The tune mts command currently emits a MTS (Non-Real Time) Single Note Tuning Change message. The number of retuned notes supported by this message seems to be limited to 127. Since one note is missing, I guess that the message is not intended to be used for whole-keyboard tunings.

If anyone knows what the correct whole-keyboard tuning strategy is, please let me know. I already tried out a bunch of other MTS messages with QSynth but with no success. Maybe I failed to set the tuning program correctly?

It should be possible to describe all current waveforms as deserializable objects.

• Run all tests
• Run cargo fmt
• Run cargo clippy

Store data in a format that is easy to edit, e.g. JSON.

Add audio-in to the Source enum s.t. it can be used as an input for any waveform.

Leave a comment if you are searching for an idea how to contribute. 😃

Access waveform buffers by a custom name and not by index

Ideas

• The Stage object continues to access waveform buffers by index
• The StageSpec object uses named buffers that need to me mapped to an index
• Use an automatically determined number of waveform buffers. The buffers are waveform-local.
• Possible extension: Use global buffers to process AudioIn or AudioOut as a whole