Skip to content

EffectCraft: An open-source motion graphics and compositing tool written by scratch in pure Rust

22.2 KB
Raw Blame
//! Audio mixdown of a composition: every audible layer (footage with audio, precomps, and any
//! layer with a sound-generating effect such as Tone) summed with its Audio Levels (dB, per
//! channel), respecting in/out points, the Audio switch and solo. Layer audio effects
//! (Effect > Audio, see [`effectcraft_effects::audio_fx`]) run on each layer's audio before
//! its levels; Backwards plays the layer's [in, out) span reversed.
//!
//! Callers mix in short blocks (one video frame, an audio device buffer, or a second of export).
//! Effect parameters are evaluated once per block. Animated Audio Levels, and animated effects
//! that work sample by sample (Stereo Mixer), are evaluated on a grid of comp samples every
//! millisecond instead, so keyframed fades ramp smoothly whatever the block size. To make the
//! result independent of block boundaries, layer effects are run from a cleared state over a
//! pre-roll of earlier audio and only the block's tail is kept — so preview playback and export
//! produce identical samples.
//!
//! Time-stretched layers are resampled nearest-sample; reversed (negative stretch) layers are
//! silent for now.

use effectcraft_effects::Params;
use effectcraft_effects::audio_fx;
use effectcraft_project::{Comp, GroupKind, ItemId, ItemKind, Layer, LayerSource, Node, Project};
use effectcraft_time::{TICKS_PER_SECOND, Tick};

use crate::{EvalCtx, ExprHost, FootageSource};

const MAX_DEPTH: usize = 16;

/// `frames` stereo sample frames of `comp`'s mix from comp time `start` at `rate` Hz,
/// interleaved (L R L R …).
pub fn mix_comp(project: &Project, footage: &dyn FootageSource, expr: Option<&dyn ExprHost>, comp: ItemId, start: Tick, frames: usize, rate: u32) -> Vec<f32> {
    let mut out = vec![0.0f32; frames * 2];
    mix_into(&AudioSources { project, footage, expr, rate }, comp, start, &mut out, 0);
    out
}

/// Whether the comp has anything audible (footage with audio, a sound generator, or a precomp
/// that has).
pub fn comp_has_audio(project: &Project, comp: ItemId) -> bool {
    fn walk(project: &Project, comp: ItemId, depth: usize) -> bool {
        let Some(c) = project.comp(comp) else { return false };
        depth < MAX_DEPTH
            && c.layers.iter().any(|l| {
                l.switches.audio
                    && (generates(l)
                        || match &l.source {
                            LayerSource::Footage { item } => {
                                matches!(project.item(*item).map(|i| &i.kind), Some(ItemKind::Footage(f)) if f.has_audio && !f.missing)
                            }
                            LayerSource::Comp { item } => walk(project, *item, depth + 1),
                            _ => false,
                        })
            })
    }
    walk(project, comp, 0)
}

/// Enabled audio effects of a layer (in stack order) with their effect ids.
fn audio_effect_groups(l: &Layer) -> impl Iterator<Item = (&str, &effectcraft_project::PropGroup)> {
    l.effects().filter(|_| l.switches.effects).into_iter().flat_map(|fx| fx.groups()).filter(|g| g.enabled).filter_map(|g| match &g.kind {
        GroupKind::Effect { effect } if audio_fx::is_audio_effect(effect) => Some((effect.as_str(), g)),
        _ => None,
    })
}

/// The layer makes sound on its own (Tone).
fn generates(l: &Layer) -> bool {
    audio_effect_groups(l).any(|(id, _)| audio_fx::generates_audio(id))
}

/// Does the layer contribute to the mix (Audio switch on and something to hear)?
pub fn audible(project: &Project, l: &Layer) -> bool {
    l.switches.audio
        && (generates(l)
            || match &l.source {
                LayerSource::Footage { item } => matches!(project.item(*item).map(|i| &i.kind), Some(ItemKind::Footage(f)) if f.has_audio),
                LayerSource::Comp { .. } => true,
                _ => false,
            })
}

/// Linear gain of an Audio Levels value in dB; -96 dB and below is silence.
pub fn db_to_gain(db: f64) -> f32 {
    if db <= -96.0 { 0.0 } else { 10f64.powf(db / 20.0) as f32 }
}

fn units_ceil(t: Tick, rate: u32) -> i64 {
    let n = t.0 as i128 * rate as i128;
    let d = TICKS_PER_SECOND as i128;
    (n.div_euclid(d) + i128::from(n.rem_euclid(d) != 0)) as i64
}

/// Samples between evaluations of animated levels and per-sample effects: one millisecond, on
/// a grid of comp sample indices so the result does not depend on where a block starts.
fn param_step(rate: u32) -> i64 {
    i64::from(rate / 1000).max(1)
}

/// Does any property of the group change over time?
fn animated(g: &effectcraft_project::PropGroup) -> bool {
    g.children.iter().any(|c| match c {
        Node::Prop(p) => p.is_animated() || p.has_expression(),
        Node::Group(sub) => animated(sub),
    })
}

fn sample_tick(i: i64, rate: u32) -> Tick {
    // Keep the requested sample when the rate does not divide the tick clock.
    let t = Tick::from_units(i, i64::from(rate));
    if t.to_units_floor(i64::from(rate)) < i { Tick(t.0.saturating_add(1)) } else { t }
}

struct MixEnv<'a> {
    project: &'a Project,
    footage: &'a dyn FootageSource,
    expr: Option<&'a dyn ExprHost>,
    comp_id: ItemId,
    comp: &'a Comp,
    rate: u32,
    depth: usize,
}

impl MixEnv<'_> {
    fn ctx(&self, t: Tick) -> EvalCtx<'_> {
        let mut ctx = EvalCtx::new(self.project, self.comp_id, self.comp, t);
        ctx.expr = self.expr;
        ctx
    }

    /// The layer's own (pre-effect) audio for comp sample indices `i0 .. i0 + n`, silent
    /// outside its [in, out) span. Comp sample index `k` maps to `in + out − 1 − k` when
    /// `reversed`.
    fn layer_raw(&self, l: &Layer, i0: i64, n: usize, reversed: bool) -> Vec<f32> {
        let rate = self.rate;
        let mut out = vec![0.0f32; n * 2];
        let (in_i, out_i) = (units_ceil(l.in_point, rate), units_ceil(l.out_point, rate));
        if reversed {
            // Mirror the clipped range, fetch it forwards, reverse.
            let a = i0.max(in_i);
            let b = (i0 + n as i64).min(out_i);
            if a >= b {
                return out;
            }
            let (ma, mb) = (in_i + out_i - b, in_i + out_i - a);
            let fwd = self.layer_raw(l, ma, (mb - ma) as usize, false);
            for k in a..b {
                let j = (in_i + out_i - 1 - k - ma) as usize;
                let o = (k - i0) as usize;
                out[o * 2] = fwd[j * 2];
                out[o * 2 + 1] = fwd[j * 2 + 1];
            }
            return out;
        }
        let a = (in_i - i0).clamp(0, n as i64) as usize;
        let b = (out_i - i0).clamp(0, n as i64) as usize;
        if a >= b {
            return out;
        }
        let m = b - a;
        let speed = 100.0 / l.stretch;
        let src_n = ((m as f64 * speed).ceil() as usize).max(1);
        let lt = l.layer_time(sample_tick(i0 + a as i64, rate));
        let buf = match &l.source {
            LayerSource::Footage { item } => match self.project.item(*item).map(|i| &i.kind) {
                Some(ItemKind::Footage(f)) if f.has_audio => self.footage.audio(*item, f, lt, src_n, rate).unwrap_or_else(|| vec![0.0; src_n * 2]),
                _ => return out,
            },
            LayerSource::Comp { item } => {
                if self.project.comp_contains(*item, self.comp_id) {
                    return out;
                }
                let mut sub = vec![0.0f32; src_n * 2];
                mix_into(&AudioSources { project: self.project, footage: self.footage, expr: self.expr, rate }, *item, lt, &mut sub, self.depth + 1);
                sub
            }
            _ => return out,
        };
        let same = (speed - 1.0).abs() < 1e-9;
        for i in 0..m {
            // The source sample at fractional index `x`, blended linearly with the next one unless the
            // layer plays at its own speed (then the samples are copied as they are).
            let x = i as f64 * speed;
            let j = if same { i } else { (x as usize).min(src_n - 1) };
            if j * 2 + 1 >= buf.len() {
                break;
            }
            let k = if same || (j + 1) * 2 + 1 >= buf.len() { j } else { j + 1 };
            let f = if same { 0.0 } else { ((x - j as f64) as f32).clamp(0.0, 1.0) };
            out[(a + i) * 2] = buf[j * 2] + (buf[k * 2] - buf[j * 2]) * f;
            out[(a + i) * 2 + 1] = buf[j * 2 + 1] + (buf[k * 2 + 1] - buf[j * 2 + 1]) * f;
        }
        out
    }

    /// Effect parameters at comp time `t`.
    fn params(&self, l: &Layer, g: &effectcraft_project::PropGroup, t: Tick) -> Params {
        let ctx = self.ctx(t);
        let mut p = Params::default();
        for c in &g.children {
            if let Node::Prop(pr) = c {
                p.values.insert(pr.match_id.clone(), ctx.value(l, pr));
            }
        }
        p
    }

    /// The layer's processed audio (effects applied, levels not) for comp sample indices
    /// `i0 .. i0 + n`.
    fn layer_audio(&self, l: &Layer, i0: i64, n: usize) -> Vec<f32> {
        let t = sample_tick(i0, self.rate);
        let fx: Vec<(&str, &effectcraft_project::PropGroup, Params)> = audio_effect_groups(l).map(|(id, g)| (id, g, self.params(l, g, t))).collect();
        if fx.is_empty() {
            return self.layer_raw(l, i0, n, false);
        }
        let reversed = fx.iter().any(|(id, _, _)| audio_fx::reverses(id));
        let pre_s: f64 = fx.iter().map(|(id, _, p)| audio_fx::preroll(id, p)).sum();
        let pre = (pre_s * self.rate as f64).ceil() as usize;
        let b0 = i0 - pre as i64;
        let mut buf = self.layer_raw(l, b0, n + pre, reversed);
        let start = l.layer_time(sample_tick(b0, self.rate)).seconds();
        for (id, g, p) in &fx {
            if audio_fx::per_sample(id) && animated(g) {
                self.process_stepped(l, id, g, &mut buf, b0);
            } else {
                audio_fx::process(id, p, &mut buf, self.rate, start);
            }
        }
        buf.drain(..pre * 2);
        buf
    }

    /// Run per-sample effect `id` over `buf` (comp samples from `b0`) in pieces of the parameter
    /// grid, each with the parameters at its first sample.
    fn process_stepped(&self, l: &Layer, id: &str, g: &effectcraft_project::PropGroup, buf: &mut [f32], b0: i64) {
        let step = param_step(self.rate);
        let frames = buf.len() / 2;
        let mut k = 0usize;
        while k < frames {
            let i = b0.saturating_add(k as i64);
            let next = i.div_euclid(step).saturating_add(1).saturating_mul(step);
            let end = usize::try_from(next.saturating_sub(b0)).unwrap_or(frames).clamp(k + 1, frames);
            let t = sample_tick(i, self.rate);
            if let Some(piece) = buf.get_mut(k * 2..end * 2) {
                audio_fx::process(id, &self.params(l, g, t), piece, self.rate, l.layer_time(t).seconds());
            }
            k = end;
        }
    }

    /// Linear gains (left, right) of the layer's Audio Levels for comp samples `i0 .. i0 + n`.
    /// Animated levels are evaluated on the parameter grid and interpolated in dB in between.
    fn level_gains(&self, l: &Layer, i0: i64, n: usize) -> Vec<[f32; 2]> {
        let Some(pr) = l.props.sub("audio").and_then(|g| g.get("levels")) else { return vec![[1.0, 1.0]; n] };
        let db = |i: i64| self.ctx(sample_tick(i, self.rate)).value(l, pr).as_vec2();
        let gain = |[a, b]: [f64; 2]| [db_to_gain(a), db_to_gain(b)];
        if !pr.is_animated() && !pr.has_expression() {
            return vec![gain(db(i0)); n];
        }
        let step = param_step(self.rate);
        let end = i0.saturating_add(n as i64);
        let mut out = Vec::with_capacity(n);
        let mut k = i0.div_euclid(step).saturating_mul(step);
        let mut lo = db(k);
        while k < end {
            let hi = db(k.saturating_add(step));
            for i in k.max(i0)..k.saturating_add(step).min(end) {
                let f = (i - k) as f64 / step as f64;
                out.push(gain([lo[0] + (hi[0] - lo[0]) * f, lo[1] + (hi[1] - lo[1]) * f]));
            }
            k = k.saturating_add(step);
            lo = hi;
        }
        out
    }
}

/// Shared inputs for composition audio mixing, including nested compositions.
struct AudioSources<'a> {
    project: &'a Project,
    footage: &'a dyn FootageSource,
    expr: Option<&'a dyn ExprHost>,
    rate: u32,
}

fn mix_into(sources: &AudioSources<'_>, comp_id: ItemId, start: Tick, out: &mut [f32], depth: usize) {
    let AudioSources { project, footage, expr, rate } = *sources;
    let Some(comp) = project.comp(comp_id) else { return };
    if depth >= MAX_DEPTH || rate == 0 {
        return;
    }
    let env = MixEnv { project, footage, expr, comp_id, comp, rate, depth };
    let frames = out.len() / 2;
    let any_solo = comp.layers.iter().any(|l| l.switches.solo && audible(project, l));
    let s0 = start.to_units_floor(rate as i64);
    for l in &comp.layers {
        if !audible(project, l) || (any_solo && !l.switches.solo) || l.stretch <= 0.0 {
            continue;
        }
        // Sample range of the block that falls inside the layer's [in, out).
        let a = (units_ceil(l.in_point, rate) - s0).clamp(0, frames as i64) as usize;
        let b = (units_ceil(l.out_point, rate) - s0).clamp(0, frames as i64) as usize;
        if a >= b {
            continue;
        }
        let gains = env.level_gains(l, s0 + a as i64, b - a);
        if gains.iter().all(|g| *g == [0.0, 0.0]) {
            continue;
        }
        let buf = env.layer_audio(l, s0 + a as i64, b - a);
        for (i, (f, g)) in buf.as_chunks::<2>().0.iter().zip(&gains).enumerate() {
            out[(a + i) * 2] += f[0] * g[0];
            out[(a + i) * 2 + 1] += f[1] * g[1];
        }
    }
}

/// A min/max peak summary of interleaved stereo audio for waveform drawing: one entry per
/// `bin` sample frames, `[min L, max L, min R, max R]`.
pub fn peak_summary(samples: &[f32], bin: usize) -> Vec<[f32; 4]> {
    let bin = bin.max(1);
    samples
        .chunks(bin * 2)
        .map(|c| {
            let mut s = [f32::MAX, f32::MIN, f32::MAX, f32::MIN];
            for f in c.as_chunks::<2>().0 {
                s[0] = s[0].min(f[0]);
                s[1] = s[1].max(f[0]);
                s[2] = s[2].min(f[1]);
                s[3] = s[3].max(f[1]);
            }
            if s[0] > s[1] { [0.0; 4] } else { s }
        })
        .collect()
}

/// Peak summary of a footage item's audio from source time 0 over `duration`, `bins_per_sec`
/// entries per second (rendered at a low internal rate, fetched in one-second chunks).
pub fn footage_peaks(footage: &dyn FootageSource, item: ItemId, f: &effectcraft_project::Footage, duration: Tick, bins_per_sec: u32) -> Vec<[f32; 4]> {
    let bins_per_sec = bins_per_sec.max(1);
    // Enough samples per bin to catch peaks of audible content, without decoding at full rate.
    let bin = (64usize).max(4000 / bins_per_sec as usize);
    let rate = bin as u32 * bins_per_sec;
    let chunk = bin * bins_per_sec as usize;
    let total = units_ceil(duration, rate).max(0) as usize;
    let mut out = Vec::with_capacity(total / bin + 1);
    let mut i = 0usize;
    while i < total {
        let n = chunk.min(total - i);
        let buf = footage.audio(item, f, sample_tick(i as i64, rate), n, rate).unwrap_or_else(|| vec![0.0; n * 2]);
        out.extend(peak_summary(&buf, bin));
        i += n;
    }
    out
}

#[cfg(test)]
mod tests {
    use std::sync::Arc;

    use effectcraft_project::{Footage, FootageKind, build};
    use effectcraft_time::FrameRate;

    use super::*;
    use crate::Image;

    /// A 1 kHz-ish constant "tone": every sample is 0.25 (left) / -0.25 (right).
    struct Dc;
    impl FootageSource for Dc {
        fn frame(&self, _: ItemId, _: &Footage, _: Tick) -> Option<Arc<Image>> {
            None
        }
        fn audio(&self, _: ItemId, _: &Footage, start: Tick, frames: usize, _: u32) -> Option<Vec<f32>> {
            Some((0..frames).flat_map(|_| if start >= Tick::ZERO { [0.25, -0.25] } else { [0.0, 0.0] }).collect())
        }
    }

    #[test]
    fn mixes_layers_with_levels_and_in_out() {
        let mut p = Project::default();
        let f = Footage {
            path: "x.wav".into(),
            kind: FootageKind::Audio,
            width: 0,
            height: 0,
            pixel_aspect: 1.0,
            frame_rate: FrameRate::new(25, 1),
            native_rate: None,
            duration: Tick::from_seconds_f64(10.0),
            has_video: false,
            has_audio: true,
            alpha: Default::default(),
            premul_color: [0.0; 3],
            loop_count: 1,
            codec: String::new(),
            missing: false,
            sequence: vec![],
            color_profile: None,
            ..Default::default()
        };
        let fid = p.add_item("x.wav", effectcraft_color::Label::SeaFoam, None, ItemKind::Footage(f));
        let comp = Comp::new(64, 64, FrameRate::new(25, 1), Tick::from_seconds_f64(2.0));
        let mut l1 = build::layer(&mut p, &comp, "a", LayerSource::Footage { item: fid }, (0, 0), None);
        l1.in_point = Tick::from_seconds_f64(1.0);
        let mut l2 = build::layer(&mut p, &comp, "b", LayerSource::Footage { item: fid }, (0, 0), None);
        if let Some(g) = l2.props.sub_mut("audio")
            && let Some(pr) = g.get_mut("levels")
        {
            pr.value = effectcraft_project::Value::Vec2([-6.0206, -96.0]);
        }
        let mut comp = comp;
        comp.layers = vec![l1, l2].into();
        let cid = p.add_item("C", effectcraft_color::Label::Sandstone, None, ItemKind::Comp(comp.into()));
        assert!(comp_has_audio(&p, cid));
        let rate = 1000;
        let m = mix_comp(&p, &Dc, None, cid, Tick::from_seconds_f64(0.5), 1000, rate);
        // first half: only layer b (half gain left, muted right)
        assert!((m[0] - 0.125).abs() < 1e-3 && m[1].abs() < 1e-6, "{} {}", m[0], m[1]);
        // second half (≥ 1 s): a + b
        let i = 700 * 2;
        assert!((m[i] - 0.375).abs() < 1e-3 && (m[i + 1] + 0.25).abs() < 1e-3, "{} {}", m[i], m[i + 1]);
    }

    /// A 440 Hz tone, so a change in the mix shows in the samples.
    struct Tone;
    impl FootageSource for Tone {
        fn frame(&self, _: ItemId, _: &Footage, _: Tick) -> Option<Arc<Image>> {
            None
        }
        fn audio(&self, _: ItemId, _: &Footage, start: Tick, frames: usize, rate: u32) -> Option<Vec<f32>> {
            let t0 = start.seconds();
            Some(
                (0..frames)
                    .flat_map(|k| {
                        let v = (std::f64::consts::TAU * 440.0 * (t0 + k as f64 / rate as f64)).sin() as f32 * 0.5;
                        [v, v]
                    })
                    .collect(),
            )
        }
    }

    #[test]
    fn a_plain_precomp_mixes_the_same_audio_as_its_layers() {
        let mut p = Project::default();
        let f = Footage {
            path: "tone.wav".into(),
            kind: FootageKind::Audio,
            has_audio: true,
            frame_rate: FrameRate::new(25, 1),
            duration: Tick::from_seconds_f64(10.0),
            ..Default::default()
        };
        let fid = p.add_item("tone.wav", effectcraft_color::Label::SeaFoam, None, ItemKind::Footage(f));
        let base = Comp::new(64, 64, FrameRate::new(25, 1), Tick::from_seconds_f64(2.0));
        let tone = build::layer(&mut p, &base, "tone", LayerSource::Footage { item: fid }, (0, 0), None);
        let mut inner = base.clone();
        inner.layers = vec![tone].into();
        let inner_id = p.add_item("Inner", effectcraft_color::Label::Sandstone, None, ItemKind::Comp(inner.into()));
        let pre = build::layer(&mut p, &base, "pre", LayerSource::Comp { item: inner_id }, (64, 64), None);
        let mut outer = base.clone();
        outer.layers = vec![pre].into();
        let outer_id = p.add_item("Outer", effectcraft_color::Label::Sandstone, None, ItemKind::Comp(outer.into()));
        let rate = 8000;
        let t = Tick::from_seconds_f64(0.5);
        let direct = mix_comp(&p, &Tone, None, inner_id, t, 4000, rate);
        let nested = mix_comp(&p, &Tone, None, outer_id, t, 4000, rate);
        assert_eq!(direct.len(), nested.len());
        let worst = direct.iter().zip(&nested).map(|(a, b)| (a - b).abs()).fold(0.0f32, f32::max);
        assert!(worst < 1e-6, "a plain precomp changes its audio by {worst}");
    }

    /// The error of a tone mixed at `stretch` percent against the exact tone, at `rate` Hz.
    fn stretch_error(stretch: f64, rate: u32) -> (f64, f64) {
        let mut p = Project::default();
        let f = Footage {
            path: "tone.wav".into(),
            kind: FootageKind::Audio,
            has_audio: true,
            frame_rate: FrameRate::new(25, 1),
            duration: Tick::from_seconds_f64(10.0),
            ..Default::default()
        };
        let fid = p.add_item("tone.wav", effectcraft_color::Label::SeaFoam, None, ItemKind::Footage(f));
        let base = Comp::new(64, 64, FrameRate::new(25, 1), Tick::from_seconds_f64(2.0));
        let mut tone = build::layer(&mut p, &base, "tone", LayerSource::Footage { item: fid }, (0, 0), None);
        tone.stretch = stretch;
        let mut comp = base.clone();
        comp.layers = vec![tone].into();
        let cid = p.add_item("C", effectcraft_color::Label::Sandstone, None, ItemKind::Comp(comp.into()));
        let n = 2000usize;
        let t = Tick::from_seconds_f64(0.5);
        let m = mix_comp(&p, &Tone, None, cid, t, n, rate);
        let speed = 100.0 / stretch;
        let (mut worst, mut sq) = (0.0f64, 0.0f64);
        for i in 0..n {
            // The source plays `speed` times as fast as the comp: source time = speed * comp time.
            let ideal = (std::f64::consts::TAU * 440.0 * speed * (t.seconds() + i as f64 / rate as f64)).sin() * 0.5;
            let e = (m[i * 2] as f64 - ideal).abs();
            worst = worst.max(e);
            sq += e * e;
        }
        (worst, (sq / n as f64).sqrt())
    }

    #[test]
    fn stretched_layers_resample_a_tone_with_small_error() {
        // Blending the two samples around each time: about 0.0002 at 48 kHz and 0.007 at 8 kHz. Copying
        // the nearest sample gave 0.025 and 0.15.
        for stretch in [70.0, 33.0] {
            let (worst, _) = stretch_error(stretch, 48_000);
            assert!(worst < 1e-3, "stretch {stretch}%: error {worst} at 48 kHz");
            let (worst, _) = stretch_error(stretch, 8_000);
            assert!(worst < 1e-2, "stretch {stretch}%: error {worst} at 8 kHz");
        }
    }
}