Skip to content

Code-first video & image, rendered natively on WebGPU.

fix(compositions): count audio layer volume keys from the layer's start

okanasl5c71a34c349c
diff --git a/.changeset/audio-volume-keys-layer-clock.md b/.changeset/audio-volume-keys-layer-clock.md
new file mode 100644
index 00000000..27e6bbc1
--- /dev/null
+++ b/.changeset/audio-volume-keys-layer-clock.md
@@ -0,0 +1,5 @@
+---
+"@framefields/compositions": patch
+---
+
+An audio layer's `volume` and `muted` keyframes count from the layer's own first frame, as a picture layer's do. They were read on the composition's clock, so any audio layer that didn't start at frame 0 was past its last key from its first sample: a fade-in `[[0, 0], [6, 1], [len - 6, 1], [len, 0]]` left every later clip silent, and keys had to be offset by `startFrame` to play. Render QA already read the keys as layer-local and flagged that workaround as `animation_truncated`; the mix and QA now agree.
diff --git a/packages/compositions/src/utils/audio-extractor.ts b/packages/compositions/src/utils/audio-extractor.ts
index 48b7b8d0..c604e8e6 100644
--- a/packages/compositions/src/utils/audio-extractor.ts
+++ b/packages/compositions/src/utils/audio-extractor.ts
@@ -687,9 +687,16 @@ export async function mixAudioTracks(
segmentStartSample + clipSamplesCount,
),
);
- // The layer's own clock: its in-point moves the source, not its animation.
+ // The layer's own clock: its in-point moves the source, not its
+ // animation, and a CompositorLayer's keyframes count from its
+ // own first frame (as its picture's do), not the composition's.
+ const layerClockFrames =
+ op.op === "CompositorLayer" ? localClockFrames : 0;
const elapsedMs =
- (totalClockFrames / fps + totalSeekSec - layerTrimSec) * 1000;
+ ((totalClockFrames - layerClockFrames) / fps +
+ totalSeekSec -
+ layerTrimSec) *
+ 1000;
const durationMs =
getActiveMediaMetadata(node)?.durationMs ?? undefined;
await pluginProcessor(slicedChannels, targetSampleRate, node, {
diff --git a/packages/framefields/src/audio-layer-timing.test.ts b/packages/framefields/src/audio-layer-timing.test.ts
index 2d12f022..222b49d6 100644
--- a/packages/framefields/src/audio-layer-timing.test.ts
+++ b/packages/framefields/src/audio-layer-timing.test.ts
@@ -2,6 +2,7 @@
* An audio layer's in-point and volume keyframes reach the mix: `trimStartSec`
* picks where the source starts (as it does for a video layer's picture), and
* a volume ramp is followed across the whole layer, not frozen at its start.
+ * Its keyframes count from the layer's own first frame, as a picture layer's do.
*/
import fs from "node:fs/promises";
import os from "node:os";
@@ -19,7 +20,10 @@ beforeAll(async () => {
const n = 2 * SR;
const ch = new Float32Array(n);
for (let i = SR; i < n; i++) ch[i] = Math.sin((2 * Math.PI * 440 * i) / SR);
- source = path.join(await fs.mkdtemp(path.join(os.tmpdir(), "audio-layer-")), "step.wav");
+ source = path.join(
+ await fs.mkdtemp(path.join(os.tmpdir(), "audio-layer-")),
+ "step.wav",
+ );
await fs.writeFile(source, encodeStereoWav([ch, ch], { sampleRate: SR }));
});
@@ -31,8 +35,8 @@ function rms(ch: Float32Array, fromSec: number, toSec: number): number {
return Math.sqrt(s / (b - a));
}
-async function mix(layer: ReturnType<typeof Layer.audio>) {
- const comp = new Composition({ width: 64, height: 64, fps: 30, durationMs: 1000 });
+async function mix(layer: ReturnType<typeof Layer.audio>, durationMs = 1000) {
+ const comp = new Composition({ width: 64, height: 64, fps: 30, durationMs });
comp.add(layer);
// renderAudio is the mix renderVideo muxes (layer processors included).
const { channels } = await new HeadlessMediaRenderer().renderAudio(
@@ -44,7 +48,9 @@ async function mix(layer: ReturnType<typeof Layer.audio>) {
describe("audio layer timing", () => {
it("starts the source at the layer's trimStartSec", async () => {
- const trimmed = await mix(Layer.audio(source, { trimStartSec: 1, durationFrames: 30 }));
+ const trimmed = await mix(
+ Layer.audio(source, { trimStartSec: 1, durationFrames: 30 }),
+ );
const untrimmed = await mix(Layer.audio(source, { durationFrames: 30 }));
// The tone begins at source 1.0 s: trimmed, it fills the layer from its
// first frame; untrimmed, the layer only hears the silence before it.
@@ -55,7 +61,10 @@ describe("audio layer timing", () => {
it("follows a volume ramp across the layer", async () => {
const ch = await mix(
Layer.audio(source, { trimStartSec: 1, durationFrames: 30 }).animate(
- LayerAnimation.keys("volume", [[0, 0], [15, 1]]),
+ LayerAnimation.keys("volume", [
+ [0, 0],
+ [15, 1],
+ ]),
),
);
const full = rms(ch, 0.6, 0.9);
@@ -66,4 +75,27 @@ describe("audio layer timing", () => {
expect(rms(ch, 0.24, 0.26)).toBeGreaterThan(full * 0.35);
expect(rms(ch, 0.24, 0.26)).toBeLessThan(full * 0.65);
});
+
+ it("counts volume keyframes from the layer's own start", async () => {
+ // The layer enters at frame 30 (1.0 s); its fade-in runs over its own
+ // first 15 frames, i.e. 1.0–1.5 s of the composition.
+ const ch = await mix(
+ Layer.audio(source, {
+ trimStartSec: 1,
+ startFrame: 30,
+ durationFrames: 30,
+ }).animate(
+ LayerAnimation.keys("volume", [
+ [0, 0],
+ [15, 1],
+ ]),
+ ),
+ 2000,
+ );
+ const full = rms(ch, 1.6, 1.9);
+ expect(full).toBeGreaterThan(0.4);
+ expect(rms(ch, 1.0, 1.03)).toBeLessThan(full * 0.1);
+ expect(rms(ch, 1.24, 1.26)).toBeGreaterThan(full * 0.35);
+ expect(rms(ch, 1.24, 1.26)).toBeLessThan(full * 0.65);
+ });
});