[readme] add info about plugins and some other stuff

README.md

@@ -10,6 +10,10 @@ as of now no plugin has ui, so they all just have a few sliders with whatever lo
 
 ## building a plugin
 
+because steinberg is an incredibly nice company, i can't actually distribute vst2 plugin binaries without a license, which they don't give out anymore :)
+
+so you're gonna have to build the plugins yourself, but don't worry cause it's relatively easy! (i think)
+
 ### macos
 
 to build a plugin, first run:
@@ -32,32 +36,32 @@ alternatively, you can run:
 ./macos.sh packagename
 ```
 
-this will generate `Packagename.vst` in `build` and copy it directly for you
+this will do all three steps directly for you
 
 ### other platforms
 
-no idea what you have to do for other platforms, refer to the `vst-rs` [docs](https://github.com/rustaudio/vst-rs). i think you can just build and take the `.so` or `.dll`, but don't quote me on that
+no idea what you have to do for other platforms, refer to the `vst-rs` [docs](https://github.com/rustaudio/vst-rs). i think you can just build with `cargo b --package package_name --release` and take the `.so` or `.dll`, but don't quote me on that
 
 PRs are welcome for more build instructions
 
-i also might delete the `build` folder from the repo,
-
 ## plugin list
 
 the following is the current list of plugins
 
-- basic_gain: simple gain plugin
-- noted: output midi at regular intervals
-- sosten: granular sustain plugin
-- quinoa: [WIP] granular synthesis plugin
-- XLowpass: reimplementation of [Airwindows XLowpass](http://www.airwindows.com/xlowpass/)
-- multidim: randomized midi note generator
-- robotuna: automatic pitch correction
-- hysteria: hysteresis nonlinear effect
-- threebandeq: 3 band eq
-- threebandwidth: 3 band stereo widener
-- tritu: say-goodbye-to-your-audio distortion
-- threebandfolding: 3 band wave folding distortion
+- `basic_gain`: simple gain plugin
+- `noted`: output midi at regular intervals
+- `sosten`: granular sustain plugin
+- `quinoa`: [WIP] granular synthesis plugin
+- `XLowpass`: reimplementation of [Airwindows XLowpass](http://www.airwindows.com/xlowpass/)
+- `multidim`: randomized midi note generator
+- `robotuna`: automatic pitch correction
+- `hysteria`: hysteresis nonlinear effect
+- `threebandeq`: 3 band eq
+- `threebandwidth`: 3 band stereo widener
+- `tritu`: say-goodbye-to-your-audio distortion
+- `threebandfolding`: 3 band wave folding distortion
+- `double_reverse_delta_inverter`: idk, a weird distortion
+- `transmute_pitch`: pitch to midi converter
 
 ### basic_gain
 
@@ -93,7 +97,7 @@ since this needs a ui so we can select a file to play, and i still don't know ho
 
 ### XLowpass
 
-XLowpass just reimplements [Airwindows XLowpass](http://www.airwindows.com/xlowpass/) in rust. it's a lowpass distortion filter
+XLowpass reimplements [Airwindows XLowpass](http://www.airwindows.com/xlowpass/) in rust. it's a lowpass distortion filter
 
 parameters:
 - `gain`
@@ -187,9 +191,31 @@ parameters:
 
 bands work as they do in `threebandeq`. the `folding freq` parameters control the frequency of the wavefolding. higher is more distortion. use values over 40 if you want to hear white noise
 
+### double-reverse delta inverter
+
+weird kinda distortion that makes loud things quiet and makes quiet things loud
+
+parameters: 
+- `dry/wet`: dry/wet control
+
+yeah there's not many params in this one, the implementation is pretty straightforward. it does weird things: a square wave of amplitude 1 will just be completely eliminated, while sine waves are distorted
+
+the effect is reversible (it's an [involution](https://en.wikipedia.org/wiki/Involution_(mathematics))), so if you add this plugin twice in a row you get the original signal back. this means you can apply drdi to your audio, apply some other effect, and apply drdi again to get some fun stuff
+
+### transmute midi
+
+analyzes pitch of incoming audio and sends out the corresponding midi note
+
+params: 
+- `passthrough`: how much of the original audio to be let through
+
+latency should be close to 1024 samples, which is around 20ms i think. the pitch detection is not excellent and it's monophonic, so if you have a noisy input or a bunch of notes at the same time it's anyone's guess what the midi output will be. there's also no pitch bending or anything fancy like that, so it'll jump around notes even if the input has portamento. 
+
+aside from that, it works well enough on my quick tests
+
 ## contributing
 
-issues and prs are welcome, but please open an issue before making any big pr, i don't wanna have to reject a pr where you have put a lot of effort on
+issues and prs are welcome, but please open an issue before making any big pr, i don't wanna have to reject a pr where you have put a lot of effort on. if you are fine with that, ig go ahead i'm not your mum
 
 ## license
 

crates/double_reverse_delta_inverter/Cargo.toml

@@ -0,0 +1,11 @@
+[package]
+name = "double_reverse_delta_inverter"
+version = "0.1.0"
+edition = "2018"
+
+[lib]
+crate-type = ["cdylib"]
+
+[dependencies]
+baseplug = { git = "https://github.com/wrl/baseplug.git", rev = "9cec68f31cca9c0c7a1448379f75d92bbbc782a8" }
+serde = "1.0.126"

crates/double_reverse_delta_inverter/src/lib.rs

@@ -0,0 +1,61 @@
+#![allow(incomplete_features)]
+#![feature(generic_associated_types)]
+
+use baseplug::{Plugin, ProcessContext};
+use serde::{Deserialize, Serialize};
+
+baseplug::model! {
+    #[derive(Debug, Serialize, Deserialize)]
+    struct InverterModel {
+        #[model(min = 0.0, max = 1.0)]
+        #[parameter(name = "dry/wet")]
+        dry_wet: f32,
+    }
+}
+
+impl Default for InverterModel {
+    fn default() -> Self {
+        Self { dry_wet: 1.0 }
+    }
+}
+
+struct Inverter;
+
+impl Plugin for Inverter {
+    const NAME: &'static str = "double-reverse delta inverter";
+    const PRODUCT: &'static str = "double-reverse delta inverter";
+    const VENDOR: &'static str = "unnieversal";
+
+    const INPUT_CHANNELS: usize = 2;
+    const OUTPUT_CHANNELS: usize = 2;
+
+    type Model = InverterModel;
+
+    #[inline]
+    fn new(_sample_rate: f32, _model: &InverterModel) -> Self {
+        Self
+    }
+
+    #[inline]
+    fn process(&mut self, model: &InverterModelProcess, ctx: &mut ProcessContext<Self>) {
+        let input = &ctx.inputs[0].buffers;
+        let output = &mut ctx.outputs[0].buffers;
+
+        for i in 0..ctx.nframes {
+            let dw = model.dry_wet[i];
+
+            output[0][i] = inv(input[0][i]) * dw + (1.0 - dw) * input[0][i];
+            output[1][i] = inv(input[1][i]) * dw + (1.0 - dw) * input[1][i];
+        }
+    }
+}
+
+// https://www.desmos.com/calculator/dbjrrevmp0
+// this is what the function does
+// it's like a mirroring of sorts
+
+fn inv(x: f32) -> f32 {
+    (1.0 - x.abs()) * x.signum()
+}
+
+baseplug::vst2!(Inverter, b"drdi");

crates/robotuna/src/lib.rs

@@ -1,19 +1,15 @@
 #![allow(incomplete_features)]
 #![feature(generic_associated_types)]
 
-use std::time::Duration;
-
 use baseplug::{MidiReceiver, Plugin, ProcessContext};
-use ringbuf::{Consumer, Producer, RingBuffer};
 use serde::{Deserialize, Serialize};
 
-use utils::buffers::*;
+use utils::delay::*;
 use utils::logs::*;
+use utils::pitch::*;
 
-mod tuna;
-
-const BUFFER_LEN: usize = 2 << 10;
-const OVERLAP: usize = BUFFER_LEN / 3;
+const BUFFER_LEN: usize = 2 << 9;
+const DELAY_LEN: usize = 4000;
 
 baseplug::model! {
     #[derive(Debug, Serialize, Deserialize)]
@@ -21,7 +17,7 @@ baseplug::model! {
         #[model(min = 0.0, max = 1.0)]
         #[parameter(name = "manual/snap")]
         manual: f32,
-        #[model(min = 0.1, max = 2.0)]
+        #[model(min = 0.1, max = 2.1)]
         #[parameter(name = "frequency gain")]
         freq_gain: f32,
     }
@@ -40,32 +36,22 @@ struct RoboTuna {
     /// Current midi note
     note: Option<u8>,
 
-    /// Current recording buffer
-    /// Input goes here
-    recording_buffer: Buffers<BUFFER_LEN>,
-    /// The next recording buffer we'll use. It gets a bit of the end of the `recording_buffer`
-    /// so we can do overlap
-    next_recording_buffer: Buffers<BUFFER_LEN>,
-    /// Current playing buffer
-    /// Output comes from here
-    playing_buffer: Buffers<BUFFER_LEN>,
-    /// Next playing buffer we'll use
-    /// We start using it at the end of the previous buffer so we can overlap
-    next_playing_buffer: Buffers<BUFFER_LEN>,
+    /// Current pitches
+    pitch_l: Option<f32>,
+    pitch_r: Option<f32>,
 
-    /// Ringbuf producer so we can send audio chunks to the processing thread
-    recordings: Producer<tuna::ProcessChunk>,
-    /// Ringbuf consumer so we can receive processed buffers from the processing threads
-    processed: Consumer<Buffers<BUFFER_LEN>>,
+    detector_thread: pitch_detection::PitchDetectorThread<BUFFER_LEN>,
 
-    /// Contains some empty buffers so we can reuse them instead of doing allocations
-    /// Buffers here are not actually empty, since we don't spend any time clearing them
-    /// But since they will be overwritten, this isn't an issue
-    empty_buffers: Vec<Buffers<BUFFER_LEN>>,
+    /// Keeps delay lines for playing
+    delays: DelayLines<DELAY_LEN>,
+
+    /// Floating indexes so we can do interpolation
+    delay_idx_l: f32,
+    delay_idx_r: f32,
+    /// true indexes so we can know how much we're drifting away
+    true_idx: usize,
 }
 
-// LMAO let's go, i think this works
-
 impl Plugin for RoboTuna {
     const NAME: &'static str = "robotuna";
     const PRODUCT: &'static str = "robotuna";
@@ -80,30 +66,23 @@ impl Plugin for RoboTuna {
     fn new(_sample_rate: f32, _model: &RoboTunaModel) -> Self {
         setup_logging("robotuna.log");
 
-        let (recordings, recording_rx) = RingBuffer::<tuna::ProcessChunk>::new(10).split();
-        let (processed_tx, processed) = RingBuffer::<Buffers<BUFFER_LEN>>::new(10).split();
-
-        // Spawn analysis thread
-        std::thread::spawn(move || {
-            tuna::tuna(recording_rx, processed_tx);
-        });
-
-        // keep some empty buffer around so we can swap them
-        let mut empty_buffers = Vec::with_capacity(50);
-        const BUF: Buffers<BUFFER_LEN> = Buffers::new();
-        empty_buffers.append(&mut vec![BUF; 30]);
+        let detector_thread = pitch_detection::PitchDetectorThread::<BUFFER_LEN>::new();
 
         log::info!("finished init");
 
         Self {
             note: None,
-            recording_buffer: Buffers::new(),
-            next_recording_buffer: Buffers::new(),
-            playing_buffer: Buffers::new(),
-            next_playing_buffer: Buffers::new(),
-            recordings,
-            processed,
-            empty_buffers,
+            pitch_l: None,
+            pitch_r: None,
+            detector_thread,
+
+            delays: DelayLines::<DELAY_LEN>::new(),
+
+            delay_idx_l: 0.0,
+            delay_idx_r: 0.0,
+            // We start this at a high number cause idk
+            // We'll catch up when we start playing
+            true_idx: 500,
         }
     }
 
@@ -113,92 +92,142 @@ impl Plugin for RoboTuna {
         let output = &mut ctx.outputs[0].buffers;
 
         for i in 0..ctx.nframes {
-            // Record
+            // pass input to pitch detector
+            self.detector_thread
+                .write(input[0][i], input[1][i], ctx.sample_rate as u32);
 
-            // Add to main buffer
-            let full = self
-                .recording_buffer
-                .write_advance(input[0][i], input[1][i]);
-            // If we're in overlap section, also add to next buffer
-            if self.recording_buffer.idx > BUFFER_LEN - OVERLAP {
-                self.next_recording_buffer
-                    .write_advance(input[0][i], input[1][i]);
-            }
-            // If we finish the buffer, switch them
-            if full {
-                // get the empty buffer from unused buffer list
-                let mut buf = self
-                    .empty_buffers
-                    .pop()
-                    .expect("should have an empty buffer");
-                buf.reset();
-                std::mem::swap(&mut buf, &mut self.recording_buffer);
-                buf.reset();
-
-                std::mem::swap(&mut self.next_recording_buffer, &mut self.recording_buffer);
-
-                let _ = self.recordings.push(tuna::ProcessChunk {
-                    buffers: buf,
-                    sample_rate: ctx.sample_rate as u32,
-                    note: self.note,
-                    manual: model.manual[i] <= 0.5,
-                    freq_gain: model.freq_gain[i],
-                });
+            // Try to get a processed buffer from the processor thread
+            if let Some((pitch_l, pitch_r)) = self.detector_thread.try_get_pitch() {
+                // Update current pitch
+                // We use `or`, so we keep the old value if the current one is None
+                self.pitch_l = pitch_l.or(self.pitch_l);
+                self.pitch_r = pitch_r.or(self.pitch_r);
             }
 
-            // Play
-
-            // Get values from main buffer
-            let (mut l, mut r, full) = self.playing_buffer.read_advance();
-            // If we're in overlap section, also play from next buffer
-            if self.playing_buffer.idx > BUFFER_LEN - OVERLAP {
-                let (l1, r1, _) = self.next_playing_buffer.read_advance();
-
-                // How much into overlap we are, from 0 to 1
-                let overlap =
-                    (OVERLAP - (BUFFER_LEN - self.playing_buffer.idx)) as f32 / OVERLAP as f32;
-
-                // Linearly crossfade
-                // lineal crossfade works well for two waves that are highly correlated
-                l *= 1. - overlap;
-                r *= 1. - overlap;
-                l += overlap * l1;
-                r += overlap * r1;
-            }
-            // If we finish the buffer, switch them
-            if full {
-                // We try to switch like with the recording buffer, but since there might not be a processed
-                // buffer yet, we do it in a loop and retry every 1 millisecond
-                // After 10 iterations we give up. Since we didn't swap the buffer, we're gonna play the last one
-                // again. This isn't ideal, but it's better than silence ig (it might not be, idk)
-
-                // The 10 iterations is arbitrary, as is the 1 millisecond wait time
-
-                for _ in 0..10 {
-                    if let Some(mut buf) = self.processed.pop() {
-                        buf.reset();
-                        std::mem::swap(&mut buf, &mut self.playing_buffer);
-                        buf.reset();
-
-                        std::mem::swap(&mut self.next_playing_buffer, &mut self.playing_buffer);
-
-                        // Stick buf in unused buffer list
-                        self.empty_buffers.push(buf);
-
-                        // Exit loop
-                        break;
-                    } else {
-                        log::info!("didn't have a processed buffer to swap to, retrying");
-                    }
-                    std::thread::sleep(Duration::from_millis(1));
-                }
-            }
+            // Play from delay line according to pitch
+            let (l, r) = self.shift(
+                input[0][i],
+                input[1][i],
+                ctx.sample_rate,
+                model.freq_gain[i],
+                model.manual[i] < 0.5,
+            );
 
             output[0][i] = l;
             output[1][i] = r;
         }
     }
 }
+impl RoboTuna {
+    fn advancement_rate(&self, freq_gain: f32, manual: bool) -> (f32, f32) {
+        // TODO Deal with pitch detection failing
+        let current_pitch_l = self.pitch_l.unwrap_or(220.0);
+        let current_pitch_r = self.pitch_r.unwrap_or(220.0);
+
+        if manual {
+            // If we're on manual, get the expected frequency from the midi note
+            if let Some(expected) = self.note.map(midi_note_to_pitch) {
+                let l = expected / current_pitch_l;
+                let r = expected / current_pitch_r;
+                (freq_gain * l, freq_gain * r)
+            } else {
+                // If there's no note, we just do frequency gain
+                (freq_gain, freq_gain)
+            }
+        } else {
+            // If we're on snap, get the closest note
+            let expected_l = closest_note_freq(current_pitch_l);
+            let expected_r = closest_note_freq(current_pitch_r);
+
+            let l = expected_l / current_pitch_l;
+            let r = expected_r / current_pitch_r;
+            (freq_gain * l, freq_gain * r)
+        }
+    }
+
+    fn shift(
+        &mut self,
+        l: f32,
+        r: f32,
+        sample_rate: f32,
+        freq_gain: f32,
+        manual: bool,
+    ) -> (f32, f32) {
+        // so um this code will probably not make any sense if i don't write an explanation of the
+        // general thing it's trying to achieve
+        // if i've forgoten to write it up and you want to understand the code, ping me and uh yeah
+
+        // add input to delay line
+        self.delays.write_and_advance(l, r);
+
+        // get period of left & right
+        let period_l = sample_rate / self.pitch_l.unwrap_or(220.0);
+        let period_r = sample_rate / self.pitch_r.unwrap_or(220.0);
+
+        // advance indexes
+        let (adv_l, adv_r) = self.advancement_rate(freq_gain, manual);
+        self.delay_idx_l += adv_l;
+        self.delay_idx_r += adv_r;
+        self.true_idx += 1;
+
+        // get how close we are to the input idx, so we know if we have to interpolate/jump
+        let l_diff = self.true_idx as f32 - self.delay_idx_l;
+        let r_diff = self.true_idx as f32 - self.delay_idx_r;
+
+        // get the current value
+        let mut l = self.delays.l.floating_index(self.delay_idx_l);
+        let mut r = self.delays.r.floating_index(self.delay_idx_r);
+
+        // Interpolation
+        // if we are close to having to jump, we start interpolating with the jump destination
+        // interpolate when we're one third of the period away from jumping
+
+        // TODO change to a non-linear interpolation
+
+        const DIV: f32 = 2.0 / 3.0;
+        if l_diff - period_l < (period_l / DIV) {
+            let a = (l_diff - period_l) / (period_l / DIV);
+            l *= a;
+            l += (1.0 - a) * self.delays.l.floating_index(self.delay_idx_l - period_l);
+        }
+        if 3.0 * period_l - l_diff < (period_l / DIV) {
+            let a = (3.0 * period_l - l_diff) / (period_l / DIV);
+            l *= a;
+            l += (1.0 - a) * self.delays.l.floating_index(self.delay_idx_l - period_l);
+        }
+        if r_diff - period_r < (period_r / DIV) {
+            let a = (r_diff - period_r) / (period_r / DIV);
+            r *= a;
+            r += (1.0 - a) * self.delays.r.floating_index(self.delay_idx_r - period_r);
+        }
+        if 3.0 * period_r - r_diff < (period_r / DIV) {
+            let a = (3.0 * period_r - r_diff) / (period_r / DIV);
+            r *= a;
+            r += (1.0 - a) * self.delays.r.floating_index(self.delay_idx_r - period_r);
+        }
+
+        // Check if we need to advance/go back `period` samples
+        // we want to be between the second and third period
+        // so ideally we want {l,r}_diff == 2.0 * period_{l,r}
+
+        // We are about to get to the first period
+        if l_diff < period_l {
+            self.delay_idx_l -= period_l;
+        }
+        // We are about to get to the fourth period
+        if l_diff > 3.0 * period_l {
+            self.delay_idx_l += period_l;
+        }
+        if r_diff < period_r {
+            self.delay_idx_r -= period_r;
+        }
+        if r_diff > 3.0 * period_r {
+            self.delay_idx_r += period_r;
+        }
+
+        (l, r)
+    }
+}
 impl MidiReceiver for RoboTuna {
     fn midi_input(&mut self, _model: &RoboTunaModelProcess, data: [u8; 3]) {
         match data[0] {

crates/robotuna/src/tuna.rs

@@ -1,131 +0,0 @@
-use ringbuf::{Consumer, Producer};
-
-use utils::buffers::*;
-use utils::pitch::*;
-
-use crate::BUFFER_LEN;
-
-type SampleRate = u32;
-
-pub struct ProcessChunk {
-    pub(crate) buffers: Buffers<BUFFER_LEN>,
-    pub(crate) sample_rate: SampleRate,
-    /// Midi note number to shift frequency to
-    pub(crate) note: Option<u8>,
-    /// If true, will listen to note
-    /// If false, will snap to closest note
-    pub(crate) manual: bool,
-    /// Extra frequency shifting to do
-    pub(crate) freq_gain: f32,
-}
-
-pub fn tuna(mut inputs: Consumer<ProcessChunk>, mut outputs: Producer<Buffers<BUFFER_LEN>>) {
-    // Keep track of last detected note, and use it in case of not detecting a new one
-    let mut prev_l_freq: Option<f32> = None;
-    let mut prev_r_freq: Option<f32> = None;
-
-    let mut detector_l = generate_pitch_detector(BUFFER_LEN);
-    let mut detector_r = generate_pitch_detector(BUFFER_LEN);
-
-    // Sample rates get overriden on first iteration, so we just set 48k
-    let mut shifter_l = generate_vocoder(48000);
-    let mut shifter_r = generate_vocoder(48000);
-
-    loop {
-        if let Some(ProcessChunk {
-            buffers: recording,
-            sample_rate,
-            note,
-            manual,
-            freq_gain,
-        }) = inputs.pop()
-        {
-            log::info!("got a buffer to process");
-
-            // If we're on manual mode, and we don't have a note, just pass through
-            if manual && note.is_none() {
-                let _ = outputs.push(recording);
-                continue;
-            }
-
-            // TODO It does weird stereo things
-
-            // Update sample rate
-            shifter_l.set_sample_rate(sample_rate as f64);
-            shifter_r.set_sample_rate(sample_rate as f64);
-
-            // Left
-            let l = recording.l;
-            // Try detecting note
-            let l = if let Some((actual, _clarity)) = pitch_detect(&mut detector_l, &l, sample_rate)
-            {
-                log::info!("L: detected actual pitch: {}", actual);
-
-                // If note is found, set it as previous, and pitch shift
-                prev_l_freq = Some(actual);
-
-                // If it's on manual mode, convert midi note to pitch
-                // If not, snap to closest frequency
-                let expected = if manual {
-                    midi_note_to_pitch(note.expect("We wouldn't be here if note is None"))
-                } else {
-                    closest_note_freq(actual)
-                };
-
-                // Perform pitch shift
-                // `expected / actual` is how much to shift the pitch
-                // If the actual pitch is 400, and expected is 800, we want to shift by 2
-                pitch_shift(&mut shifter_l, &l, freq_gain * expected / actual)
-            } else if let Some(actual) = prev_l_freq {
-                log::info!("L: reusing actual pitch: {}", actual);
-
-                let expected = if manual {
-                    midi_note_to_pitch(note.expect("We wouldn't be here if note is None"))
-                } else {
-                    closest_note_freq(actual)
-                };
-
-                pitch_shift(&mut shifter_l, &l, freq_gain * expected / actual)
-            } else {
-                log::info!("L: no actual pitch");
-
-                // If there's nothing, leave it as is
-                l
-            };
-
-            // Same thing for the right side
-            let r = recording.r;
-            let r = if let Some((actual, _clarity)) = pitch_detect(&mut detector_r, &r, sample_rate)
-            {
-                log::info!("R: detected actual pitch: {}", actual);
-
-                prev_r_freq = Some(actual);
-
-                let expected = if manual {
-                    midi_note_to_pitch(note.expect("We wouldn't be here if note is None"))
-                } else {
-                    closest_note_freq(actual)
-                };
-
-                pitch_shift(&mut shifter_r, &l, freq_gain * expected / actual)
-            } else if let Some(actual) = prev_r_freq {
-                log::info!("R: reusing actual pitch: {}", actual);
-
-                let expected = if manual {
-                    midi_note_to_pitch(note.expect("We wouldn't be here if note is None"))
-                } else {
-                    closest_note_freq(actual)
-                };
-
-                pitch_shift(&mut shifter_r, &l, freq_gain * expected / actual)
-            } else {
-                log::info!("R: no actual pitch");
-
-                r
-            };
-
-            let _ = outputs.push(Buffers::from(l, r));
-            log::info!("finished processing a buffer");
-        }
-    }
-}

crates/sosten/Cargo.toml

@@ -9,3 +9,5 @@ crate-type = ["cdylib"]
 [dependencies]
 baseplug = { git = "https://github.com/wrl/baseplug.git", rev = "9cec68f31cca9c0c7a1448379f75d92bbbc782a8" }
 serde = "1.0.126"
+
+utils = { path = "../utils" }

crates/sosten/src/delay.rs

@@ -1,30 +0,0 @@
-pub struct DelayLine<const LEN: usize> {
-    buffer: [f32; LEN],
-    index: usize,
-}
-
-impl<const LEN: usize> DelayLine<LEN> {
-    pub fn new() -> Self {
-        Self {
-            buffer: [0.0; LEN],
-            index: 0,
-        }
-    }
-
-    pub fn read_slice(&self, slice: &mut [f32]) {
-        // Copy values in order
-        for i in 0..LEN {
-            slice[i] = self.buffer[(self.index + i - LEN) % LEN];
-        }
-    }
-
-    pub fn write_and_advance(&mut self, value: f32) {
-        self.buffer[self.index] = value;
-
-        if self.index == LEN - 1 {
-            self.index = 0;
-        } else {
-            self.index += 1;
-        }
-    }
-}

crates/sosten/src/lib.rs

@@ -4,8 +4,7 @@
 use baseplug::{Plugin, ProcessContext};
 use serde::{Deserialize, Serialize};
 
-mod delay;
-use delay::DelayLine;
+use utils::delay::*;
 
 // If you change this remember to change the max on the model
 const LEN: usize = 48000;

crates/transmute_pitch/Cargo.toml

@@ -0,0 +1,15 @@
+[package]
+name = "transmute_pitch"
+version = "0.1.0"
+edition = "2018"
+
+[lib]
+crate-type = ["cdylib"]
+
+[dependencies]
+baseplug = { git = "https://github.com/wrl/baseplug.git", rev = "9cec68f31cca9c0c7a1448379f75d92bbbc782a8" }
+serde = "1.0.126"
+log = "0.4.14"
+ringbuf = "0.2.5"
+
+utils = { path = "../utils" }

crates/transmute_pitch/src/lib.rs

@@ -0,0 +1,108 @@
+#![allow(incomplete_features)]
+#![feature(generic_associated_types)]
+
+use baseplug::{event::Data, Event, Plugin, ProcessContext};
+use serde::{Deserialize, Serialize};
+
+use utils::pitch::*;
+
+const BUFFER_LEN: usize = 2 << 9;
+
+baseplug::model! {
+    #[derive(Debug, Serialize, Deserialize)]
+    struct TransmutePitchModel {
+        #[model(min = 0.0, max = 1.0)]
+        #[parameter(name = "passthrough")]
+        passthrough: f32,
+    }
+}
+
+impl Default for TransmutePitchModel {
+    fn default() -> Self {
+        Self { passthrough: 1.0 }
+    }
+}
+
+struct TransmutePitch {
+    detector_thread: pitch_detection::PitchDetectorThread<BUFFER_LEN>,
+    last_note: Option<u8>,
+}
+
+impl Plugin for TransmutePitch {
+    const NAME: &'static str = "transmute pitch";
+    const PRODUCT: &'static str = "transmute pitch";
+    const VENDOR: &'static str = "unnieversal";
+
+    const INPUT_CHANNELS: usize = 1;
+    const OUTPUT_CHANNELS: usize = 1;
+
+    type Model = TransmutePitchModel;
+
+    #[inline]
+    fn new(_sample_rate: f32, _model: &TransmutePitchModel) -> Self {
+        let detector_thread = pitch_detection::PitchDetectorThread::<BUFFER_LEN>::new();
+        Self {
+            detector_thread,
+            last_note: None,
+        }
+    }
+
+    #[inline]
+    fn process(&mut self, model: &TransmutePitchModelProcess, ctx: &mut ProcessContext<Self>) {
+        let input = &ctx.inputs[0].buffers;
+        let output = &mut ctx.outputs[0].buffers;
+        let enqueue_midi = &mut ctx.enqueue_event;
+
+        for i in 0..ctx.nframes {
+            output[0][i] = model.passthrough[i] * input[0][i];
+            output[1][i] = model.passthrough[i] * input[1][i];
+
+            // pass input to pitch detector
+            self.detector_thread
+                .write(input[0][i], 0.0, ctx.sample_rate as u32);
+
+            // Try to get a processed buffer from the processor thread
+            match self.detector_thread.try_get_pitch() {
+                Some((Some(pitch), _)) => {
+                    let note = pitch_to_midi_note(pitch);
+
+                    // If note changed
+                    if self.last_note != Some(note) {
+                        // Send note off for last note
+                        if let Some(last_note) = self.last_note {
+                            let note_off = Event::<TransmutePitch> {
+                                frame: i,
+                                data: Data::Midi([0x80, last_note, 0]),
+                            };
+                            enqueue_midi(note_off);
+                        }
+
+                        // Send note on for the new one
+                        let note_on = Event::<TransmutePitch> {
+                            frame: i,
+                            data: Data::Midi([0x90, note, 64]),
+                        };
+                        enqueue_midi(note_on);
+
+                        // Update note
+                        self.last_note = Some(note);
+                    }
+                }
+                Some((None, _)) => {
+                    if let Some(last_note) = self.last_note {
+                        let note_off = Event::<TransmutePitch> {
+                            frame: i,
+                            data: Data::Midi([0x80, last_note, 0]),
+                        };
+                        enqueue_midi(note_off);
+                    }
+
+                    self.last_note = None;
+                }
+                _ => {}
+            }
+        }
+    }
+}
+
+baseplug::vst2!(TransmutePitch, b"trpi");

crates/utils/Cargo.toml

@@ -10,3 +10,4 @@ log = "0.4.14"
 log-panics = "2.0.0"
 dirs = "3.0.2"
 pvoc = { path = "../pvoc-rs" }
+ringbuf = "0.2.5"

crates/utils/src/delay.rs

@@ -0,0 +1,88 @@
+pub struct DelayLine<const LEN: usize> {
+    buffer: [f32; LEN],
+    index: usize,
+}
+impl<const LEN: usize> DelayLine<LEN> {
+    pub fn new() -> Self {
+        Self {
+            buffer: [0.0; LEN],
+            index: 0,
+        }
+    }
+
+    pub fn read_slice(&self, slice: &mut [f32]) {
+        // Copy values in order
+        for i in 0..LEN {
+            slice[i] = self.wrapped_index(self.index + i);
+        }
+    }
+
+    pub fn write_and_advance(&mut self, value: f32) {
+        self.buffer[self.index] = value;
+
+        if self.index == LEN - 1 {
+            self.index = 0;
+        } else {
+            self.index += 1;
+        }
+    }
+
+    /// Returns the sample at idx after taking modulo LEN
+    pub fn wrapped_index(&self, idx: usize) -> f32 {
+        self.buffer[idx % LEN]
+    }
+
+    /// Indexes the buffer but interpolates between the current and the next sample
+    pub fn floating_index(&self, val: f32) -> f32 {
+        let idx = val.trunc() as usize;
+        let frac = val.fract();
+
+        // TODO uhm idk what this should be, but we don't want an underflow so yeah,
+        let xm1 = if idx == 0 {
+            0.0
+        } else {
+            self.wrapped_index(idx - 1)
+        };
+        let x0 = self.wrapped_index(idx);
+        let x1 = self.wrapped_index(idx + 1);
+        let x2 = self.wrapped_index(idx + 2);
+
+        // linear interpolation
+        // return (1.0 - frac) * x0 + frac * x1;
+
+        crate::hermite(frac, xm1, x0, x1, x2)
+    }
+
+    /// Get a reference to the delay line's index.
+    pub fn idx(&self) -> &usize {
+        &self.index
+    }
+
+    /// Get a reference to the delay line's buffer.
+    pub fn buffer(&self) -> &[f32; LEN] {
+        &self.buffer
+    }
+}
+
+pub struct DelayLines<const LEN: usize> {
+    pub l: DelayLine<LEN>,
+    pub r: DelayLine<LEN>,
+}
+impl<const LEN: usize> DelayLines<LEN> {
+    pub fn new() -> Self {
+        Self {
+            l: DelayLine::<LEN>::new(),
+            r: DelayLine::<LEN>::new(),
+        }
+    }
+
+    pub fn read_slices(&self, l: &mut [f32], r: &mut [f32]) {
+        self.l.read_slice(l);
+        self.r.read_slice(r);
+    }
+
+    pub fn write_and_advance(&mut self, l: f32, r: f32) {
+        self.l.write_and_advance(l);
+        self.r.write_and_advance(r);
+    }
+}

crates/utils/src/lib.rs

@@ -1,4 +1,15 @@
 pub mod buffers;
+pub mod delay;
 pub mod logs;
 pub mod pitch;
 pub mod threeband;
+
+pub fn hermite(frac: f32, xm1: f32, x0: f32, x1: f32, x2: f32) -> f32 {
+    let c = (x1 - xm1) * 0.5;
+    let v = x0 - x1;
+    let w = c + v;
+    let a = w + v + (x2 - x0) * 0.5;
+    let b_neg = w + a;
+
+    (((a * frac) - b_neg) * frac + c) * frac + x0
+}

crates/utils/src/pitch/mod.rs

@@ -1,30 +1,4 @@
-use pitch_detection::detector::yin::YINDetector;
-use pitch_detection::detector::PitchDetector;
-
-pub fn generate_pitch_detector(size: usize) -> impl PitchDetector<f32> {
-    let padding = size / 2;
-
-    YINDetector::new(size, padding)
-}
-
-/// Returns an option with (Frequency, Clarity)
-pub fn pitch_detect(
-    detector: &mut dyn PitchDetector<f32>,
-    signal: &[f32],
-    sample_rate: u32,
-) -> Option<(f32, f32)> {
-    const POWER_THRESHOLD: f32 = 0.15;
-    const CLARITY_THRESHOLD: f32 = 0.5;
-
-    let pitch = detector.get_pitch(
-        &signal,
-        sample_rate as usize,
-        POWER_THRESHOLD,
-        CLARITY_THRESHOLD,
-    );
-
-    pitch.map(|a| (a.frequency, a.clarity))
-}
+pub mod pitch_detection;
 
 pub fn generate_vocoder(sample_rate: u32) -> PhaseVocoder {
     PhaseVocoder::new(1, sample_rate as f64, 256, 4)

crates/utils/src/pitch/pitch_detection.rs

@@ -0,0 +1,144 @@
+use pitch_detection::detector::yin::YINDetector;
+use pitch_detection::detector::PitchDetector;
+
+use crate::buffers::Buffers;
+use ringbuf::{Consumer, Producer, RingBuffer};
+
+pub fn generate_pitch_detector(size: usize) -> impl PitchDetector<f32> {
+    let padding = size / 2;
+
+    YINDetector::new(size, padding)
+}
+
+/// Returns an option with (Frequency, Clarity)
+pub fn pitch_detect(
+    detector: &mut dyn PitchDetector<f32>,
+    signal: &[f32],
+    sample_rate: u32,
+) -> Option<(f32, f32)> {
+    const POWER_THRESHOLD: f32 = 0.15;
+    const CLARITY_THRESHOLD: f32 = 0.5;
+
+    let pitch = detector.get_pitch(
+        &signal,
+        sample_rate as usize,
+        POWER_THRESHOLD,
+        CLARITY_THRESHOLD,
+    );
+
+    pitch.map(|a| (a.frequency, a.clarity))
+}
+
+pub struct DetectionInput<const LEN: usize> {
+    pub buffers: Buffers<LEN>,
+    pub sample_rate: u32,
+}
+
+pub struct DetectionOutput<const LEN: usize> {
+    pub buffers: Buffers<LEN>,
+    pub pitch_l: Option<f32>,
+    pub pitch_r: Option<f32>,
+}
+
+pub fn detect<const LEN: usize>(
+    mut inputs: Consumer<DetectionInput<LEN>>,
+    mut outputs: Producer<DetectionOutput<LEN>>,
+) {
+    let mut detector_l = generate_pitch_detector(LEN);
+    let mut detector_r = generate_pitch_detector(LEN);
+
+    loop {
+        if let Some(DetectionInput::<LEN> {
+            buffers,
+            sample_rate,
+        }) = inputs.pop()
+        {
+            let pitch_l = pitch_detect(&mut detector_l, &buffers.l, sample_rate).map(|a| a.0);
+            let pitch_r = pitch_detect(&mut detector_r, &buffers.r, sample_rate).map(|a| a.0);
+
+            let _ = outputs.push(DetectionOutput {
+                buffers,
+                pitch_l,
+                pitch_r,
+            });
+        }
+    }
+}
+
+pub struct PitchDetectorThread<const LEN: usize> {
+    /// Current recording buffer
+    /// Input goes here
+    recording_buffer: Buffers<LEN>,
+
+    /// Ringbuf producer so we can send audio chunks to the processing thread
+    recordings: Producer<DetectionInput<LEN>>,
+    /// Ringbuf consumer so we can receive processed buffers from the processing threads
+    processed: Consumer<DetectionOutput<LEN>>,
+
+    /// Contains some empty buffers so we can reuse them instead of doing allocations
+    /// Buffers here are not actually empty, since we don't spend any time clearing them
+    /// But since they will be overwritten, this isn't an issue
+    empty_buffers: Vec<Buffers<LEN>>,
+}
+
+impl<const LEN: usize> PitchDetectorThread<LEN> {
+    pub fn new() -> Self {
+        let (recordings, recording_rx) = RingBuffer::<DetectionInput<LEN>>::new(30).split();
+        let (processed_tx, processed) = RingBuffer::<DetectionOutput<LEN>>::new(30).split();
+
+        // Spawn analysis thread
+        std::thread::spawn(move || {
+            detect(recording_rx, processed_tx);
+        });
+
+        // keep some empty buffer around so we can swap them
+        let mut empty_buffers = Vec::with_capacity(80);
+        empty_buffers.append(&mut vec![Buffers::new(); 30]);
+
+        Self {
+            recordings,
+            processed,
+            empty_buffers,
+            recording_buffer: Buffers::new(),
+        }
+    }
+
+    pub fn write(&mut self, l: f32, r: f32, sample_rate: u32) {
+        let full = self.recording_buffer.write_advance(l, r);
+        // If we fill the buffer, switch it with an empty one
+        if full {
+            // we have to loop here, cause when the daw renders audio it tries to do it faster than
+            // real time. if we don't loop and wait, the processing thread gets stuck with all of the buffers,
+            // and we run out of empty ones to switch to
+            // the loop-wait ensures that we don't panic when there isn't an empty buffer
+            loop {
+                // get the empty buffer from unused buffer list
+                if let Some(mut buf) = self.empty_buffers.pop() {
+                    buf.reset();
+                    // swap it with recording buffer
+                    std::mem::swap(&mut buf, &mut self.recording_buffer);
+                    buf.reset();
+
+                    // pass it to the processor thread
+                    let _ = self.recordings.push(DetectionInput::<LEN> {
+                        buffers: buf,
+                        sample_rate,
+                    });
+                    break;
+                }
+                std::thread::sleep(std::time::Duration::from_micros(10));
+            }
+        }
+    }
+
+    pub fn try_get_pitch(&mut self) -> Option<(Option<f32>, Option<f32>)> {
+        let DetectionOutput::<LEN> {
+            buffers,
+            pitch_l,
+            pitch_r,
+        } = self.processed.pop()?;
+        self.empty_buffers.push(buffers);
+
+        Some((pitch_l, pitch_r))
+    }
+}